paper_id stringlengths 9 12 | venue stringclasses 139
values | year stringclasses 7
values | paper_title stringlengths 0 181 | paper_authors stringlengths 4 925 | paper_abstract stringlengths 1 5k | paper_keywords stringlengths 2 436 | paper_content stringlengths 0 100k | review_id stringlengths 9 12 | review_title stringlengths 0 500 | review_rating stringclasses 61
values | review_text stringlengths 2 28.3k | review_confidence stringclasses 13
values | text stringlengths 402 130k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
r1xX42R5Fm | ICLR.cc/2019/Conference | 2019 | Beyond Greedy Ranking: Slate Optimization via List-CVAE | ["Ray Jiang", "Sven Gowal", "Yuqiu Qian", "Timothy Mann", "Danilo J. Rezende"] | The conventional approach to solving the recommendation problem greedily ranks
individual document candidates by prediction scores. However, this method fails to
optimize the slate as a whole, and hence, often struggles to capture biases caused
by the page layout and document interdepedencies. The slate recommendation
... | ["CVAE", "VAE", "recommendation system", "slate optimization", "whole page optimization"] | ABSTRACTThe conventional solution to the recommendation problem greedily ranks individualdocument candidates by prediction scores. However, this method fails to optimizethe slate as a whole, and hence, often struggles to capture biases caused by thepage layout and document interdepedencies. The slate recommendation pro... | BklYOAG62m | Using variational auto-encoders to go beyond greedy construction of slates for recommendations | 6: Marginally above acceptance threshold | The latest revision is a substantially improved version of the paper. The comment about generalization still feels unsatisfying ("our model requires choosing c* in the support of P(c) seen during training") but could spur follow-up work attempting a precise characterization.
I remain wary of using a neural net reward f... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Beyond Greedy Ranking: Slate Optimization via List-CVAE
### Paper Abstract
The conventional approach to solving the recommendation problem greedily ranks individual docu... |
SWNM52GXh0Y | ML_Reproducibility_Challenge/2021/Fall | 2021 | [Reproducibility Report] Explainable Deep One-Class Classification | ["Anonymous"] | Scope of Reproducibility
Liznerski et al. [23] proposed Fully Convolutional Data Description (FCDD), an explainable version of the Hypersphere
Classifier (HSC) to directly address image anomaly detection (AD) and pixel-wise AD without any post-hoc explainer
methods. The authors claim that FCDD achieves results compara... | ["Reproducibility", "Deep Learning", "Explainable AI (XAI)", "Anomaly Detection", "Anomaly Segmentation", "Pixel-wise Anomaly Detection"] | [Reproducibility Report] Explainable Deep One-ClassClassificationAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2Liznerski et al. [ 23] proposed Fully Convolutional Data Description (FCDD), an explainable version of the Hypersphere 3Classifier (HSC) to directly address image... | Hcef6bEol5 | RE:[Reproducibility Report] Explainable Deep One-Class Classification | 6: Marginally above acceptance threshold | Summary:
This paper presents a report on the reproducibility of the fully Convolutional Data Description(FCDD) method for image anomaly detection(AD). This study verifies comparable performance of FCDD on fashion-MNIST and CIFAR-10 and outperforms existing methods on MVTec-AD on the pixel-wise task. However, additional... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
[Reproducibility Report] Explainable Deep One-Class Classification
### Paper Abstract
Scope of Reproducibility Liznerski et al. [23] proposed Fully Convolutional Data D... |
b1tl3aOt2R2 | robot-learning.org/CoRL/2023/Conference | 2023 | GNFactor: Multi-Task Real Robot Learning with Generalizable Neural Feature Fields | ["Yanjie Ze", "Ge Yan", "Yueh-Hua Wu", "Annabella Macaluso", "Yuying Ge", "Jianglong Ye", "Nicklas Hansen", "Li Erran Li", "Xiaolong Wang"] | It is a long-standing problem in robotics to develop agents capable of executing diverse manipulation tasks from visual observations in unstructured real-world environments. To achieve this goal, the robot will need to have a comprehensive understanding of the 3D structure and semantics of the scene. In this work, we ... | ["Robotic Manipulation", "Neural Radiance Field", "Behavior Cloning"] | GNFactor: Multi-Task Real Robot Learning withGeneralizable Neural Feature FieldsYanjie Ze1∗Ge Yan2∗Yueh-Hua Wu2∗Annabella Macaluso2Yuying Ge3Jianglong Ye2Nicklas Hansen2Li Erran Li4Xiaolong Wang21Shanghai Jiao Tong University2UC San Diego3University of Hong Kong4AWS AI, Amazon∗Equal Contributionyanjieze.com/GNFactorAbs... | TQSdTAI7Hp4 | The biggest strength and main contribution of this paper is the way in which the voxel encoder is trained. In addition to a loss on the output Q-values, a secondary loss on the reconstruction of the scene is used. Specifically, the encoded voxel grid is interpreted as a neural radiance field and is re-rendered from ano... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
GNFactor: Multi-Task Real Robot Learning with Generalizable Neural Feature Fields
### Paper Abstract
It is a long-standing problem in robotics to develop agents capable ... | ||
bodgPrarPUJ | ICLR.cc/2021/Conference | 2021 | Lipschitz-Bounded Equilibrium Networks | ["Max Revay", "Ruigang Wang", "Ian Manchester"] | This paper introduces new parameterizations of equilibrium neural networks, i.e.
networks defined by implicit equations. This model class includes standard multilayer
and residual networks as special cases. The new parameterization admits
a Lipschitz bound during training via unconstrained optimization, i.e. no project... | ["Adversarial Robustness", "Equilibrium Networks", "Neural ODE"] | ABSTRACTThis paper introduces new parameterizations of equilibrium neural networks, i.e.networks defined by implicit equations. This model class includes standard mul-tilayer and residual networks as special cases. The new parameterization admitsa Lipschitz bound during training via unconstrained optimization: no proje... | C1MrdKsPHbK | Solid theoretical results that generalized previous formulations; empirical results could be improved | 7: Good paper, accept | > Summary: This paper studies a new and more general way of parameterizing the simplest equilibrium network of the form $\sigma(Wz+Ux+b)$, a form that has been tackled by works like (Winston & Kolter 2020)and (El Ghaoui et al. 2019). The authors provide a computationally (relatively) efficient way of computing Lipschit... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Lipschitz-Bounded Equilibrium Networks
### Paper Abstract
This paper introduces new parameterizations of equilibrium neural networks, i.e. networks defined by implicit e... |
Skl-fAVYvH | ICLR.cc/2020/Conference | 2020 | One Demonstration Imitation Learning | ["Bradly C. Stadie", "Siyan Zhao", "Qiqi Xu", "Bonnie Li", "Lunjun Zhang"] | We develop a new algorithm for imitation learning from a single expert demonstration. In contrast to many previous one-shot imitation learning approaches, our algorithm does not assume access to more than one expert demonstration during the training phase. Instead, we leverage an exploration policy to acquire unsupervi... | ["imitation learning", "one shot imitation learning", "reinforcement learning", "exploration", "representation learning"] | r1lhh7aKtS | Official Blind Review #1 | 3: Weak Reject | This paper studies the problem of one-shot imitation learning. The aim is to learn a policy which, given a single expert demonstration, can mimic the expert's behavior. Making progress on this problem is important for decreasing the amount of human supervision required to teach intelligent agents. This paper proposes t... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
One Demonstration Imitation Learning
### Paper Abstract
We develop a new algorithm for imitation learning from a single expert demonstration. In contrast to many previou... | ||
SkJdByaUG | ICLR.cc/2018/Workshop | 2018 | Forecasting probability distribution of nonlinear time series | ["Kyongmin Yeo", "Igor Melnyk", "Nam Nguyen", "Eun Kyung Lee"] | We propose DE-RNN to learn the probability density function (PDF) of a nonlinear time series, and compute the temporal evolution of the PDF for a probabilistic forecast. A Recurrent Neural Network (RNN) based model is employed to learn a nonlinear operator for temporal evolution of the stochastic process. We use a soft... | ["Recurrent Neural Network", "forecast", "density estimation", "nonlinear time series", "dynamical system", "uncertainty quantification"] | ABSTRACTWe propose DE-RNN to learn the probability density function (PDF) of a nonlin-ear time series, and compute the temporal evolution of the PDF for a probabilisticforecast. A Recurrent Neural Network (RNN) based model is employed to learn anonlinear operator for temporal evolution of the stochastic process. We use... | H1fjkOgtM | Well-known combination of LSTM and discretisation of time series | 5: Marginally below acceptance threshold | This paper studies the use of LSTMs to do nonlinear time-series forecasting while explicitly expressing prediction uncertainty, by modeling the univariate or multivariate scalar values as probability distribution functions using multinomials over discretised time-series.
Following an LSTM by a softmax and replacing re... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Forecasting probability distribution of nonlinear time series
### Paper Abstract
We propose DE-RNN to learn the probability density function (PDF) of a nonlinear time se... |
B1eYGkBKDB | ICLR.cc/2020/Conference | 2020 | Fully Quantized Transformer for Improved Translation | ["Gabriele Prato", "Ella Charlaix", "Mehdi Rezagholizadeh"] | State-of-the-art neural machine translation methods employ massive amounts of parameters. Drastically reducing computational costs of such methods without affecting performance has been up to this point unsolved. In this work, we propose a quantization strategy tailored to the Transformer architecture. We evaluate our ... | ["Transformer", "quantization", "machine translation", "compression", "pruning"] | ABSTRACTState-of-the-art neural machine translation methods employ massive amounts ofparameters. Drastically reducing computational costs of such methods withoutaffecting performance has been up to this point unsuccessful. To the best of ourknowledge, we are the first to propose a quantization strategy inclusive of all... | SyghRPs15S | Official Blind Review #2 | 1: Reject | This paper proposes a quantization strategy for Transformer models and demonstrate results on a neural machine translation benchmark. Although the results look promising, the experiments are a bit sparse in terms of details. In my opinion, an empirical performance-focused paper like this really needs to provide a lot o... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Fully Quantized Transformer for Improved Translation
### Paper Abstract
State-of-the-art neural machine translation methods employ massive amounts of parameters. Drastic... | |
B1GRtj05t7 | ICLR.cc/2019/Conference | 2019 | NA | ["NA"] | NA | [] | r1luSeas2X | Review | 6: Marginally above acceptance threshold | This paper described a spatial value decomposition for RL. The idea is: instead of temporally sum reward for Q function, multiply future states will be visited at the same time. Each future state is weighted by visiting frequency. The new Q function will be weighted average of future states reward. In order to make th... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NA
### Paper Abstract
NA
### Paper Keywords
[]
### Paper Content
<|im_end|>
<|im_start|>assistant
### Review Title
Review
### Review Text
This paper described a spati... | |
4rsTcjH7co | ICLR.cc/2021/Conference | 2021 | Autoencoder Image Interpolation by Shaping the Latent Space | ["Alon Oring", "Zohar Yakhini", "Yacov Hel-Or"] | One of the fascinating properties of deep learning is the ability of the network to reveal the underlying factors characterizing elements in datasets of different types. Autoencoders represent an effective approach for computing these factors. Autoencoders have been studied in the context of enabling interpolation betw... | ["deep learning", "autoencoders", "deep generative models", "representation learning"] | ABSTRACTAutoencoders represent an effective approach for computing the underlying factorscharacterizing datasets of different types. The latent representation of autoencodershave been studied in the context of enabling interpolation between data points bydecoding convex combinations of latent vectors. This interpolatio... | eE4IIJeaYCa | Interpolating Autoencoder | 5: Marginally below acceptance threshold | This paper focused on developing a new regularization technique for autoencoders, which shapes the latent representation to follow a manifold that is consistent with the training images and that drives the manifold to be smooth and locally convex. The authors suggest that the manifold structure of continuous data must ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Autoencoder Image Interpolation by Shaping the Latent Space
### Paper Abstract
One of the fascinating properties of deep learning is the ability of the network to reveal... |
YTyHkF4P03w | ICLR.cc/2021/Conference | 2021 | What to Prune and What Not to Prune at Initialization | ["Maham Haroon"] | Post-training dropout based approaches achieve high sparsity and are well established means of deciphering problems relating to computational cost and overfitting in Neural Network architectures. Contrastingly, pruning at initialization is still far behind. Initialization pruning is more efficacious when it comes to sc... | ["Network Sparsity", "Machine Learning", "Initialization Pruning"] | ABSTRACTPost-training dropout based approaches achieve high sparsity and are well estab-lished means of deciphering problems relating to computational cost and overfit-ting in Neural Network architectures citesrivastava2014dropout, (Pan et al., 2016),Zhu & Gupta (2017), LeCun et al. (1990). Contrastingly, pruning at in... | 2nByxB4bQkz | A sketch of suggestions with no serious intention | 1: Trivial or wrong | Summary: The authors propose two approaches for pruning: (a) "Evolution-style": start with K random masks associated with the weights, update weights on gradient descent corresponding to those active in the “fittest” mask, and overtime throw away all but one masks which are less fit. (b) "Dissipating-gradients”: Here t... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
What to Prune and What Not to Prune at Initialization
### Paper Abstract
Post-training dropout based approaches achieve high sparsity and are well established means of d... |
CSqjS121nsU | ICML.cc/2020/Workshop/SAS | 2020 | ASR Free End-to-End Spoken Language Understanding using Transformers | ["Karan Purohit", "Deepesh Bhageria", "Vishwa Nath Jha"] | End-to-end spoken language understanding (SLU) systems directly map speech to intent through a single trainable model whereas conventional SLU systems use Automatic Speech Recognition (ASR) to convert speech to text and utilize Natural Language Understanding (NLU) to get intent. In this paper, we show how transformer-b... | ["Transformer", "Spoken Language Understanding"] | 000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044045046047048049050051052053054ASR free End-to-End SLU using the TransformerAnonymous Authors1AbstractEnd-to-end spoken language understanding (SLU)systems directly map speech to intent ... | 0HahNocQP2e | A transformer-based end-to-end architecture for spoken language understanding (SLU) | 4: Ok but not good enough - rejection | A transformer-based end-to-end architecture for spoken language understanding (SLU) is proposed for detecting
intents composed of three concepts: action, object, and location in the Fluent Speech Command dataset.
quality
The task has a limited semantic content mentioned in a complex acoustic environment.
Concept me... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ASR Free End-to-End Spoken Language Understanding using Transformers
### Paper Abstract
End-to-end spoken language understanding (SLU) systems directly map speech to int... |
rkeiQlBFPB | ICLR.cc/2020/Conference | 2020 | Meta-Learning with Warped Gradient Descent | ["Sebastian Flennerhag", "Andrei A. Rusu", "Razvan Pascanu", "Francesco Visin", "Hujun Yin", "Raia Hadsell"] | Learning an efficient update rule from data that promotes rapid learning of new tasks from the same distribution remains an open problem in meta-learning. Typically, previous works have approached this issue either by attempting to train a neural network that directly produces updates or by attempting to learn better i... | ["meta-learning", "transfer learning"] | ABSTRACTLearning an efficient update rule from data that promotes rapid learning of newtasks from the same distribution remains an open problem in meta-learning. Typ-ically, previous works have approached this issue either by attempting to traina neural network that directly produces updates or by attempting to learn b... | HkgHvrYptS | Official Blind Review #1 | 8: Accept | This paper proposes a learning strategy to precondition gradients for meta-learning. I really enjoyed reading the paper though I admit that I couldn't fully grasp all the details yet (paper is dense). My comments below are mostly to improve the readability of the paper for readers like me (knowing a thing or two in opt... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Meta-Learning with Warped Gradient Descent
### Paper Abstract
Learning an efficient update rule from data that promotes rapid learning of new tasks from the same distrib... | |
HyxjOyrKvr | ICLR.cc/2020/Conference | 2020 | Neural Epitome Search for Architecture-Agnostic Network Compression | ["Daquan Zhou", "Xiaojie Jin", "Qibin Hou", "Kaixin Wang", "Jianchao Yang", "Jiashi Feng"] | Traditional compression methods including network pruning, quantization, low rank factorization and knowledge distillation all assume that network architectures and parameters should be hardwired. In this work, we propose a new perspective on network compression, i.e., network parameters can be disentangled from the a... | ["Network Compression", "Classification", "Deep Learning", "Weights Sharing"] | ABSTRACTTraditional compression methods including network pruning, quantization, lowrank factorization and knowledge distillation all assume that network architecturesand parameters are one-to-one mapped. In this work, we propose a new perspectiveon network compression, i.e., network parameters can be disentangled from... | S1lORAAAcB | Official Blind Review #3 | 6: Weak Accept | This paper focuses on the problem of neural network compression, and proposes a new scheme, the neural epitome search. It learns to find compact yet expressive epitomes for weight parameters of a specified network architecture. The learned weight tensors are independent of the architecture design. It can be encapsula... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Epitome Search for Architecture-Agnostic Network Compression
### Paper Abstract
Traditional compression methods including network pruning, quantization, low rank ... | |
Twf5rUVeU-I | ICLR.cc/2021/Conference | 2021 | Convergence Analysis of Homotopy-SGD for Non-Convex Optimization | ["Matilde Gargiani", "Andrea Zanelli", "Moritz Diehl", "Quoc Tran-Dinh", "Frank Hutter"] | First-order stochastic methods for solving large-scale non-convex optimization problems are widely used in many big-data applications, e.g. training deep neural networks as well as other complex and potentially non-convex machine learning
models. Their inexpensive iterations generally come together with slow global con... | ["deep learning", "numerical optimization", "transfer learning"] | ABSTRACTFirst-order stochastic methods for solving large-scale non-convex optimizationproblems are widely used in many big-data applications, e.g. training deep neuralnetworks as well as other complex and potentially non-convex machine learningmodels. Their inexpensive iterations generally come together with slow globa... | AGqSNDYzB2A | This algorithm is not that practical and the the hypothesis is a bit vague. | 4: Ok but not good enough - rejection | 1. It seems to me the proposed Homotopy-SGD is not a practical algorithm, as in each iteration the algorithm has to solve a nontrivial (possibly nonconvex) subproblem. In other words, each subproblem can be as difficult as the original problem. This leads to an essential question that what is the practical motivation o... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Convergence Analysis of Homotopy-SGD for Non-Convex Optimization
### Paper Abstract
First-order stochastic methods for solving large-scale non-convex optimization proble... |
uIc4W6MtbDA | ICLR.cc/2021/Conference | 2021 | ERMAS: Learning Policies Robust to Reality Gaps in Multi-Agent Simulations | ["Eric Zhao", "Alexander R Trott", "Caiming Xiong", "Stephan Zheng"] | Policies for real-world multi-agent problems, such as optimal taxation, can be learned in multi-agent simulations with AI agents that emulate humans. However, simulations can suffer from reality gaps as humans often act suboptimally or optimize for different objectives (i.e., bounded rationality). We introduce $\epsilo... | ["Robustness", "Multi-Agent Learning", "Sim2Real", "Reinforcement Learning"] | ABSTRACTPolicies for real-world multi-agent problems, such as optimal taxation, can belearned in multi-agent simulations with AI agents that emulate humans. How-ever, simulations can suffer from reality gaps as humans often act suboptimallyor optimize for different objectives (i.e., bounded rationality). We introduce -... | Pzos_xIApLT | Great motivation, some minor qualms on execution | 7: Good paper, accept | This paper proposes a method to learn _robust_ policies in multiagent environments: in particular, the policies should continue to work even if other agents in the environment deviate slightly in their behavior. Unlike previous work which allows for perturbations in other agents’ policies, this work allows for perturba... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ERMAS: Learning Policies Robust to Reality Gaps in Multi-Agent Simulations
### Paper Abstract
Policies for real-world multi-agent problems, such as optimal taxation, can... |
HkpbnH9lx | ICLR.cc/2017/conference | 2017 | Density estimation using Real NVP | ["Laurent Dinh", "Jascha Sohl-Dickstein", "Samy Bengio"] | Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning. Specifically, designing models with tractable learning, sampling, inference and evaluation is crucial in solving this task. We extend the space of such models using real-valued non-volume preserving (real NVP) transf... | ["Deep learning", "Unsupervised Learning"] | ABSTRACTUnsupervised learning of probabilistic models is a central yet challenging problemin machine learning. Specifically, designing models with tractable learning, sam-pling, inference and evaluation is crucial in solving this task. We extend the spaceof such models using real-valued non-volume preserving (real NVP)... | ryV5E2-4e | 7: Good paper, accept | Building on earlier work on a model called NICE, this paper presents an approach to constructing deep feed-forward generative models. The model is evaluated on several datasets. While it does not achieve state-of-the-art performance, it advances an interesting class of models. The paper is mostly well written and clear... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Density estimation using Real NVP
### Paper Abstract
Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning. Specifically... | |
HJ_aoCyRZ | ICLR.cc/2018/Conference | 2018 | SpectralNet: Spectral Clustering using Deep Neural Networks | ["Uri Shaham", "Kelly Stanton", "Henry Li", "Ronen Basri", "Boaz Nadler", "Yuval Kluger"] | Spectral clustering is a leading and popular technique in unsupervised data analysis. Two of its major limitations are scalability and generalization of the spectral embedding (i.e., out-of-sample-extension). In this paper we introduce a deep learning approach to spectral clustering that overcomes the above shortcomin... | ["unsupervised learning", "spectral clustering", "siamese networks"] | ABSTRACTSpectral clustering is a leading and popular technique in unsupervised data anal-ysis. Two of its major limitations are scalability and generalization of the spec-tral embedding (i.e., out-of-sample-extension). In this paper we introduce a deeplearning approach to spectral clustering that overcomes the above sh... | HylbmO3lf | SpectralNet: Spectral Clustering using Deep Neural Networks | 7: Good paper, accept | Brief Summary:
The paper introduces a deep learning based approach that approximates spectral clustering. The basic idea is to train a neural network to map a representation of input in the eigenspace where k-means clustering can be performed as usual. The method is more scalable than the vanilla spectral clustering al... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
SpectralNet: Spectral Clustering using Deep Neural Networks
### Paper Abstract
Spectral clustering is a leading and popular technique in unsupervised data analysis. Two... |
L8BElg6Qldb | ICLR.cc/2021/Conference | 2021 | Nonvacuous Loss Bounds with Fast Rates for Neural Networks via Conditional Information Measures | ["Fredrik Hellstr\u00f6m", "Giuseppe Durisi"] | We present a framework to derive bounds on the test loss of randomized learning algorithms for the case of bounded loss functions. This framework leads to bounds that depend on the conditional information density between the the output hypothesis and the choice of the training set, given a larger set of data samples fr... | ["bounds", "training set", "fast rates", "neural networks", "framework", "conditional information density", "literature", "nonvacuous loss"] | ABSTRACTWe present a framework to derive bounds on the test loss of randomized learningalgorithms for the case of bounded loss functions. This framework leads to boundsthat depend on the conditional information density between the the output hypothe-sis and the choice of the training set, given a larger set of data sam... | 8f32MMQq8-c | Steady theoretical progress while a little unsatisfactory validation | 6: Marginally above acceptance threshold | This paper derives bounds on the test loss under the random-subset setting, which are expressed with conditional information measures, and have fast rates with respect to the sample size n. The derived bounds are compared with the practical performance of deep neural networks (DNNs) trained on the MNIST and Fashion-MNI... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Nonvacuous Loss Bounds with Fast Rates for Neural Networks via Conditional Information Measures
### Paper Abstract
We present a framework to derive bounds on the test lo... |
r-VN8ru89gc | ICLR.cc/2022/Workshop/OSC | 2022 | Object-centric Generative Models for Spatial Scene Understanding | ["Anonymous"] | Representing a scene and its constituent objects from raw sensory data is a core ability for enabling robots to interact with their environment. In this paper, we propose a novel system for scene understanding, leveraging object-centric generative models. We demonstrate an agent that is able to learn and reason about 3... | ["generative models", "spatial scene", "agent", "able", "scene", "constituent objects", "raw sensory data", "core ability", "robots", "environment"] | ABSTRACTRepresenting a scene and its constituent objects from raw sensory data is a coreability for enabling robots to interact with their environment. In this paper, wepropose a novel system for scene understanding, leveraging object-centric gener-ative models. We demonstrate an agent that is able to learn and reason ... | BdWe1G5c3Wq | Object-centric Generative Models for Spatial Scene Understanding | 2: Good workshop paper, accept | Summary:
This work builds on Van de Maele et al. [1] who introduce Cortical Column Networks [CCNs]. A CCN encodes an image patch containing an object into an object ID latent variable and a pose latent variable. The pose latent variable can be combined with an action to compute a transition which allows an object to be... | 2: The reviewer is somewhat confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Object-centric Generative Models for Spatial Scene Understanding
### Paper Abstract
Representing a scene and its constituent objects from raw sensory data is a core abil... |
Bke03G85DN | ICLR.cc/2019/Workshop/drlStructPred | 2019 | Robust Reinforcement Learning for Autonomous Driving | ["Yesmina Jaafra", "Jean Luc Laurent", "Aline Deruyver", "Mohamed Saber Naceur"] | Autonomous driving is still considered as an “unsolved problem” given its inherent important variability and that many processes associated with its development like vehicle control and scenes recognition remain open issues. Despite reinforcement learning algorithms have achieved notable results in games and some robot... | ["Neural networks", "Deep reinforcement learning", "Actor-critic model", "Autonomous driving", "Carla simulator"] | ABSTRACTAutonomous driving is still considered as an “unsolved problem” given its inher-ent important variability and that many processes associated with its developmentlike vehicle control and scenes recognition remain open issues. Despite reinforce-ment learning algorithms have achieved notable results in games and s... | HJg-dj07tE | Nice task but it should have more experiments. | 3: Marginally above acceptance threshold | This paper evaluates two common RL algorithms in a simulated driving environment.
Pros:
1- Nice choice of using a more realistic environment.
2- The paper is easy to read and understand.
Cons:
1- Some details are missing: how exactly the reward is computed? How long an episode lasts, on average? What is the average d... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust Reinforcement Learning for Autonomous Driving
### Paper Abstract
Autonomous driving is still considered as an “unsolved problem” given its inherent important var... |
SkliR1SKDS | ICLR.cc/2020/Conference | 2020 | Rethinking Data Augmentation: Self-Supervision and Self-Distillation | ["Hankook Lee", "Sung Ju Hwang", "Jinwoo Shin"] | Data augmentation techniques, e.g., flipping or cropping, which systematically enlarge the training dataset by explicitly generating more training samples, are effective in improving the generalization performance of deep neural networks. In the supervised setting, a common practice for data augmentation is to assign t... | ["self-supervision", "data augmentation"] | ABSTRACTData augmentation techniques, e.g., flipping or cropping, which systematicallyenlarge the training dataset by explicitly generating more training samples, areeffective in improving the generalization performance of deep neural networks.In the supervised setting, a common practice for data augmentation is to ass... | BJeySVo6KH | Official Blind Review #2 | 3: Weak Reject | This paper proposes to use data-augmentation as self-supervision, applied as a regularizer in fully supervised classification settings in order to improve the generalization accuracy of the classifier. It uses a joint (label x augmentation) prediction space, rather than a factorized space of labels and augmentations in... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Rethinking Data Augmentation: Self-Supervision and Self-Distillation
### Paper Abstract
Data augmentation techniques, e.g., flipping or cropping, which systematically en... | |
ByetGn0cYX | ICLR.cc/2019/Conference | 2019 | Probabilistic Planning with Sequential Monte Carlo methods | ["Alexandre Piche", "Valentin Thomas", "Cyril Ibrahim", "Yoshua Bengio", "Chris Pal"] | In this work, we propose a novel formulation of planning which views it as a probabilistic inference problem over future optimal trajectories. This enables us to use sampling methods, and thus, tackle planning in continuous domains using a fixed computational budget. We design a new algorithm, Sequential Monte Carlo... | ["control as inference", "probabilistic planning", "sequential monte carlo", "model based reinforcement learning"] | ABSTRACTIn this work, we propose a novel formulation of planning which views it as aprobabilistic inference problem over future optimal trajectories. This enables usto use sampling methods, and thus, tackle planning in continuous domains usinga fixed computational budget. We design a new algorithm, Sequential MonteCarl... | SylcRv_DhQ | Interesting preliminary work, but requires major revisions | 5: Marginally below acceptance threshold | The authors formulate planning as sampling from an intractable distribution motivated by control-as-inference, propose to approximately sample from the distribution using a learned model of the environment and SMC, then evaluate their approach on 3 Mujoco tasks. They claim that their method compares favorably to model-... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Probabilistic Planning with Sequential Monte Carlo methods
### Paper Abstract
In this work, we propose a novel formulation of planning which views it as a probabilistic ... |
xfOVXyO_cwJ | ICLR.cc/2021/Conference | 2021 | Empirical Frequentist Coverage of Deep Learning Uncertainty Quantification Procedures | ["Benjamin Kompa", "Jasper Snoek", "Andrew Beam"] | Uncertainty quantification for complex deep learning models is increasingly important as these techniques see growing use in high-stakes, real-world settings. Currently, the quality of a model's uncertainty is evaluated using point-prediction metrics such as negative log-likelihood or the Brier score on heldout data. I... | ["uncertainty quantification", "coverage", "dataset shift"] | ABSTRACTUncertainty quantification for complex deep learning models is increasingly im-portant as these techniques see growing use in high-stakes, real-world settings.Currently, the quality of a model’s uncertainty is evaluated using point-predictionmetrics such as negative log-likelihood or the Brier score on heldout ... | EexjCtRnul | Confused definition of coverage | 3: Clear rejection | In the submitted manuscript, "Empirical Frequentist coverage of deep learning uncertainty quantification procedures", the authors propose to investigate the Frequentist coverage properties of predictive intervals by numerical experiment for a number of machine learning models applied to benchmark datasets. I can't say... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Empirical Frequentist Coverage of Deep Learning Uncertainty Quantification Procedures
### Paper Abstract
Uncertainty quantification for complex deep learning models is i... |
Syl7OsRqY7 | ICLR.cc/2019/Conference | 2019 | Coarse-grain Fine-grain Coattention Network for Multi-evidence Question Answering | ["Victor Zhong", "Caiming Xiong", "Nitish Shirish Keskar", "Richard Socher"] | End-to-end neural models have made significant progress in question answering, however recent studies show that these models implicitly assume that the answer and evidence appear close together in a single document. In this work, we propose the Coarse-grain Fine-grain Coattention Network (CFC), a new question answering... | ["question answering", "reading comprehension", "nlp", "natural language processing", "attention", "representation learning"] | ABSTRACTEnd-to-end neural models have made significant progress in question answering,however recent studies show that these models implicitly assume that the answerand evidence appear close together in a single document. In this work, we proposethe Coarse-grain Fine-grain Coattention Network (CFC), a new question answ... | HkxVLCtKnm | Method that is well adapted to the task to be solved; clear and well written | 7: Good paper, accept | This paper proposes a method for multi-hop QA based on two separate modules, which are called coarse-grained and fine-grained modules. The coarse-grained module reads all of the supporting documents for QA, whereas the fine-grained one reads the local context surrounding each candidate entity's mentions. Both modules a... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Coarse-grain Fine-grain Coattention Network for Multi-evidence Question Answering
### Paper Abstract
End-to-end neural models have made significant progress in question ... |
Sks9_ajex | ICLR.cc/2017/conference | 2017 | Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer | ["Sergey Zagoruyko", "Nikos Komodakis"] | Attention plays a critical role in human visual experience. Furthermore, it has recently been demonstrated that attention can also play an important role in the context of applying artificial neural networks to a variety of tasks from fields such as computer vision and NLP. In this work we show that, by properly defini... | ["Computer vision", "Deep learning", "Supervised Learning"] | ABSTRACTAttention plays a critical role in human visual experience. Furthermore, it hasrecently been demonstrated that attention can also play an important role in thecontext of applying artificial neural networks to a variety of tasks from fields suchas computer vision and NLP. In this work we show that, by properly d... | HyiC5efNg | Some nice results, but it is not clear what are the advantages/drawbacks of the different attention maps | 6: Marginally above acceptance threshold | This paper proposes to investigate attention transfers between a teacher and a student network.
Attention transfer is performed by minimising the l2 distance between the teacher/student attention maps at different layers, in addition to minimising the classification loss and optionally a knowledge distillation term.
... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer
### Paper Abstract
Attention plays a critical role ... |
Byl5NREFDr | ICLR.cc/2020/Conference | 2020 | Thieves on Sesame Street! Model Extraction of BERT-based APIs | ["Kalpesh Krishna", "Gaurav Singh Tomar", "Ankur P. Parikh", "Nicolas Papernot", "Mohit Iyyer"] | We study the problem of model extraction in natural language processing, in which an adversary with only query access to a victim model attempts to reconstruct a local copy of that model. Assuming that both the adversary and victim model fine-tune a large pretrained language model such as BERT (Devlin et al., 2019), we... | ["model extraction", "BERT", "natural language processing", "pretraining language models", "model stealing", "deep learning security"] | ABSTRACTWe study the problem of model extraction in natural language processing, inwhich an adversary with only query access to a victim model attempts to recon-struct a local copy of that model. Assuming that both the adversary and victimmodel fine-tune a large pretrained language model such as BERT (Devlin et al.,201... | SJlh2oSRKH | Official Blind Review #1 | 8: Accept | This authors introduce a novel approach to successful modern extraction. The paper is well written and easy to follow (the two exceptions/oddities are Figure 1 & Table 1, which appear one page before they are refered, which makes them initially hard to understand because they are out of context). The experimental eval... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Thieves on Sesame Street! Model Extraction of BERT-based APIs
### Paper Abstract
We study the problem of model extraction in natural language processing, in which an adv... | |
SkVhlh09tX | ICLR.cc/2019/Conference | 2019 | Pay Less Attention with Lightweight and Dynamic Convolutions | ["Felix Wu", "Angela Fan", "Alexei Baevski", "Yann Dauphin", "Michael Auli"] | Self-attention is a useful mechanism to build generative models for language and images. It determines the importance of context elements by comparing each element to the current time step. In this paper, we show that a very lightweight convolution can perform competitively to the best reported self-attention results. ... | ["Deep learning", "sequence to sequence learning", "convolutional neural networks", "generative models"] | ABSTRACTSelf-attention is a useful mechanism to build generative models for language andimages. It determines the importance of context elements by comparing each ele-ment to the current time step. In this paper, we show that a very lightweight convo-lution can perform competitively to the best reported self-attention ... | H1gJoj8C3X | Major advance in sequence-to-sequence architectures | 8: Top 50% of accepted papers, clear accept | The authors present lightweight convolutions and dynamic convolutions, two significant advances over existing depthwise convolution sequence models, and demonstrate very strong results on machine translation, language modeling, and summarization. Their results go even further than those of the Transformer paper in coun... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Pay Less Attention with Lightweight and Dynamic Convolutions
### Paper Abstract
Self-attention is a useful mechanism to build generative models for language and images. ... |
ryBnUWb0b | ICLR.cc/2018/Conference | 2018 | Predicting Floor-Level for 911 Calls with Neural Networks and Smartphone Sensor Data | ["William Falcon", "Henning Schulzrinne"] | In cities with tall buildings, emergency responders need an accurate floor level location to find 911 callers quickly. We introduce a system to estimate a victim's floor level via their mobile device's sensor data in a two-step process. First, we train a neural network to determine when a smartphone enters or exits a b... | ["Recurrent Neural Networks", "RNN", "LSTM", "Mobile Device", "Sensors"] | ABSTRACTIn cities with tall buildings, emergency responders need an accurate floor levellocation to find 911 callers quickly. We introduce a system to estimate a victim’sfloor level via their mobile device’s sensor data in a two-step process. First, wetrain a neural network to determine when a smartphone enters or exit... | ryca0nYef | The paper combines existing methods to outperform baseline methods on floor level estimation. Limitations of their approach are not explored. | 6: Marginally above acceptance threshold | The authors motivate the problem of floor level estimation and tackle it with a RNN. The results are good. The models the authors compare to are well chosen. As the paper foremost provides application (and combination) of existing methods it would be benefitial to know something about the limitations of their approach ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Predicting Floor-Level for 911 Calls with Neural Networks and Smartphone Sensor Data
### Paper Abstract
In cities with tall buildings, emergency responders need an accur... |
hb-FRSqrNTU | MICCAI.org/2021/Challenge/KiTS | 2021 | Transformer-based 3D U-Net with attention blocks for Kidney Tumor Segmentation | ["Bohdan Petryshak", "Dmytro Fishman", "Danylo Kolinko", "Igor Krashenyi"] | Here we describe our envisioned solution to The 2021 Kidney and Kidney Tumor Segmentation Challenge which constitutes training the original U-Net architecture with a wide range of modifications, including transformers, residual connections, and attention blocks. Once quantitative results are obtained we are happy to mo... | ["segmentation", "attention block", "transformer"] | Transformer-based 3D U-Net with attentionblocks for Kidney Tumor SegmentationBohdan Petryshak1;2;3, Danylo Kolinko3, Igor Krashenyi4, and DmytroFishman1;21Better Medicine O U, Tartu, Estonia2Department of Computer Science, University of Tartu, Narva mnt. 18, 51009,Estonia3Ukrainian Catholic University, Ilariona Svjents... | -jX3rtam_mj | Official Review | 4: Ok but not good enough - rejection | ### Overall
- The paper makes explicit mention of being unfinished. This is fine for the original submission but please be sure to remove these statements in your revision
### Introduction
- Please cite your sources for the epidemiology numbers and clarify whether they are worldwide or pertain only to a particular c... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Transformer-based 3D U-Net with attention blocks for Kidney Tumor Segmentation
### Paper Abstract
Here we describe our envisioned solution to The 2021 Kidney and Kidney ... | |
B1eO9oA5Km | ICLR.cc/2019/Conference | 2019 | A Guider Network for Multi-Dual Learning | ["Wenpeng Hu", "Zhengwei Tao", "Zhanxing Zhu", "Bing Liu", "Zhou Lin", "Jinwen Ma", "Dongyan Zhao", "Rui Yan"] | A large amount of parallel data is needed to train a strong neural machine translation (NMT) system. This is a major challenge for low-resource languages. Building on recent work on unsupervised and semi-supervised methods, we propose a multi-dual learning framework to improve the performance of NMT by using an almost ... | ["guider network", "parallel data", "languages", "large amount", "system", "major challenge", "building", "recent work", "unsupervised"] | ABSTRACTA large amount of parallel data is needed to train a strong neural machine translation(NMT) system. This is a major challenge for low-resource languages. Building onrecent work on unsupervised and semi-supervised methods, we propose a multi-dual learning framework to improve the performance of NMT by using an a... | HkxHWLGthm | Not easy to follow; experiments not convincing | 4: Ok but not good enough - rejection | [Summary]
This paper proposes an extension of the dual learning framework, with a guider network and multiple languages included: (1) Each language $i$ has a guider network $GN_i$, that can be used to reconstruct the source sentence from either the output of the encoder or the output of the decoder. (2) Multiple langua... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Guider Network for Multi-Dual Learning
### Paper Abstract
A large amount of parallel data is needed to train a strong neural machine translation (NMT) system. This is ... |
BMua55nUyyt | ICLR.cc/2021/Conference | 2021 | Median DC for Sign Recovery: Privacy can be Achieved by Deterministic Algorithms | ["Jiyuan Tu", "Weidong Liu", "Xiaojun Mao"] | Privacy-preserving data analysis becomes prevailing in recent years. It is a common sense in privacy literature that strict differential privacy can only be obtained by imposing additional randomness in the algorithm. In this paper, we study the problem of private sign recovery for sparse mean estimation and sparse lin... | ["Median-of-means", "divide-and-conquer", "privacy", "sign recovery"] | ABSTRACTPrivacy-preserving data analysis becomes prevailing in recent years. It is a com-mon sense in privacy literature that strict differential privacy can only be obtainedby imposing additional randomness in the algorithm. In this paper, we study theproblem of private sign recovery for sparse mean estimation and spa... | -rcll8AVSLQ | Median DC review. Decision: Reject. | 4: Ok but not good enough - rejection | The paper gives "almost private" algorithms for problem of sign recovery of mean vector and of linear regression. The techniques follow their general framework of Median DC, which is similar to the well-known median-of-means approach. They give theoretical guarantees for the same, along with empirical results comparing... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Median DC for Sign Recovery: Privacy can be Achieved by Deterministic Algorithms
### Paper Abstract
Privacy-preserving data analysis becomes prevailing in recent years. ... |
FovoQL3nygw | ACM.org/ICMI/2023/Workshop/GENEA_Challenge | 2023 | FEIN-Z: Autoregressive Behavior Cloning for Speech-Driven Gesture Generation | ["Leon Harz", "Hendric Vo\u00df", "Stefan Kopp"] | Human communication relies on multiple modalities such as verbal expressions, facial cues, and bodily gestures. Developing computational approaches to process and generate these multimodal signals is critical for seamless human-agent interaction. A particular challenge is the generation of co-speech gestures due to the... | ["machine learning", "deep learning", "co-speech gesture generation", "gesture synthesis", "multimodal data", "transformer", "behavior cloning", "reinforcement learning"] | ABSTRACTHuman communication relies on multiple modalities such as verbalexpressions, facial cues, and bodily gestures. Developing compu-tational approaches to process and generate these multimodal sig-nals is critical for seamless human-agent interaction. A particularchallenge is the generation of co-speech gestures du... | eaiY-C2JvOi | An interesting paper | 8: Top 50% of accepted papers, clear accept | [Paper Summary]
This paper proposed a co-speech generation framework combining transformer and Wasserstein GAN architecture. The model was inspired by BC-Z network architecture, and the authors replaced the vision network component in BC-Z with an attention-based network. The model takes audio, text, and speaker iden... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
FEIN-Z: Autoregressive Behavior Cloning for Speech-Driven Gesture Generation
### Paper Abstract
Human communication relies on multiple modalities such as verbal expressi... |
B1ZvaaeAZ | ICLR.cc/2018/Conference | 2018 | WRPN: Wide Reduced-Precision Networks | ["Asit Mishra", "Eriko Nurvitadhi", "Jeffrey J Cook", "Debbie Marr"] | For computer vision applications, prior works have shown the efficacy of reducing numeric precision of model parameters (network weights) in deep neural networks. Activation maps, however, occupy a large memory footprint during both the training and inference step when using mini-batches of inputs. One way to reduce th... | ["Low precision", "binary", "ternary", "4-bits networks"] | ABSTRACTFor computer vision applications, prior works have shown the efficacy of re-ducing numeric precision of model parameters (network weights) in deep neu-ral networks. Activation maps, however, occupy a large memory footprint dur-ing both the training and inference step when using mini-batches of inputs. Oneway to... | HJiml81-z | Review of WRPN | 5: Marginally below acceptance threshold | This paper presents an simple and interesting idea to improve the performance for neural nets. The idea is we can reduce the precision for activations and increase the number of filters, and is able to achieve better memory usage (reduced). The paper is aiming to solve a practical problem, and has done some solid resea... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
WRPN: Wide Reduced-Precision Networks
### Paper Abstract
For computer vision applications, prior works have shown the efficacy of reducing numeric precision of model par... |
SJxfxnA9K7 | ICLR.cc/2019/Conference | 2019 | Structured Prediction using cGANs with Fusion Discriminator | ["Faisal Mahmood", "Wenhao Xu", "Nicholas J. Durr", "Jeremiah W. Johnson", "Alan Yuille"] | We propose a novel method for incorporating conditional information into a generative adversarial network (GAN) for structured prediction tasks. This method is based on fusing features from the generated and conditional information in feature space and allows the discriminator to better capture higher-order statistics ... | ["Generative Adversarial Networks", "GANs", "conditional GANs", "Discriminator", "Fusion"] | ABSTRACTWe propose a novel method for incorporating conditional information into a gen-erative adversarial network (GAN) for structured prediction tasks. This method isbased on fusing features from the generated and conditional information in featurespace and allows the discriminator to better capture higher-order stat... | SkgLoB3p3Q | This paper presents a new method for incorporating conditional information into a GAN for structured prediction tasks (image conditioned GAN problems). Thorough experimental results on Cityscapes and NYU v2 verify the efficacy of the proposed method. | 5: Marginally below acceptance threshold | This paper presents a new method for incorporating conditional information into a GAN for structured prediction tasks (image conditioned GAN problems). The proposed method is based on fusing features from the generated and conditional information in feature space and allows the discriminator to better capture higher-or... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Structured Prediction using cGANs with Fusion Discriminator
### Paper Abstract
We propose a novel method for incorporating conditional information into a generative adve... |
SkwAEQbAb | ICLR.cc/2018/Conference | 2018 | A novel method to determine the number of latent dimensions with SVD | ["Asana Neishabouri", "Michel Desmarais"] | Determining the number of latent dimensions is a ubiquitous problem in machine
learning. In this study, we introduce a novel method that relies on SVD to discover
the number of latent dimensions. The general principle behind the method is to
compare the curve of singular values of the SVD decomposition of a data set wi... | ["SVD", "Latent Dimensions", "Dimension Reductions", "Machine Learning"] | ABSTRACTDetermining the number of latent dimensions is a ubiquitous problem in machinelearning. In this study, we introduce a novel method that relies on SVD to discoverthe number of latent dimensions. The general principle behind the method is tocompare the curve of singular values of the SVD decomposition of a data s... | r1N3gmtlz | Not ready for publication | 2: Strong rejection | The authors propose a bootstrap-based test for determining the number of latent dimensions to retain for linear dimensionality reduction (SVD/PCA). The idea is to retain eigenvectors which are larger than a bootstrap average. The resulting approach is evaluated on two simulated datasets (dense and sparse)as compared to... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A novel method to determine the number of latent dimensions with SVD
### Paper Abstract
Determining the number of latent dimensions is a ubiquitous problem in machine le... |
7xArdn_FKtV | ICLR.cc/2021/Conference | 2021 | Heterogeneous Model Transfer between Different Neural Networks | ["Guangcong Wang", "Jianhuang Lai", "Wenqi Liang", "Guangrun Wang"] | We propose an effective heterogeneous model transfer (HMT) method that can transfer the knowledge from one pretrained neural network to another neural network. Most of the existing deep learning methods depend much on a pretraining-finetuning strategy, i.e., pretraining a deep model on a large task-related (source) dat... | ["Heterogeneous model transfer", "pretraining-finetuning"] | ABSTRACTWe propose an effective heterogeneous model transfer (HMT) method that cantransfer the knowledge from one pretrained neural network to another neural net-work. Most of the existing deep learning methods depend much on a pretraining-finetuning strategy, i.e., pretraining a deep model on a large task-related (sou... | ykrVNoicw_s | review | 5: Marginally below acceptance threshold | This paper proposes an interesting model transfer setting: transfer knowledge from one pretrained neural network to another neural network with different network architecture. The challenges of this transfer problem are: (1) how to find the layer-to-layer relationships and (2) how to transfer weights between two layers... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Heterogeneous Model Transfer between Different Neural Networks
### Paper Abstract
We propose an effective heterogeneous model transfer (HMT) method that can transfer the... |
BJxoz1rKwr | ICLR.cc/2020/Conference | 2020 | CRAP: Semi-supervised Learning via Conditional Rotation Angle Prediction | ["Hai-Ming Xu", "Lingqiao Liu"] | Self-supervised learning (SlfSL), aiming at learning feature representations through ingeniously designed pretext tasks without human annotation, has achieved compelling progress in the past few years. Very recently, SlfSL has also been identified as a promising solution for semi-supervised learning (SemSL) since it of... | ["Semi-supervised Learning", "Self-supervised Learning"] | ABSTRACTSelf-supervised learning (SlfSL), aiming at learning feature representationsthrough ingeniously designed pretext tasks without human annotation, hasachieved compelling progress in the past few years. Very recently, SlfSL hasalso been identified as a promising solution for semi-supervised learning (SemSL)since i... | r1ewToY0YS | Official Blind Review #3 | 3: Weak Reject | This paper proposed a framework to solve semi-supervised learning problems using self-supervised learning, By implementing it through rotation angle prediction, called CRAP. It predicts the image rotationangle conditioned on the candidate image class. Experimental evaluations show that CRAP achieves superior performan... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
CRAP: Semi-supervised Learning via Conditional Rotation Angle Prediction
### Paper Abstract
Self-supervised learning (SlfSL), aiming at learning feature representations ... | |
EwsLcX5NRKr | ICLR.cc/2021/Conference | 2021 | Deep Active Learning for Object Detection with Mixture Density Networks | ["Jiwoong Choi", "Ismail Elezi", "Hyuk-Jae Lee", "Clement Farabet", "Jose M. Alvarez"] | Active learning aims to reduce the labeling costs by selecting only samples that are informative to improve the accuracy of the network. Few existing works have addressed the problem of active learning for object detection, and most of them estimate the informativeness of an image based only on the classification head,... | ["Active Learning", "Object Detection"] | ABSTRACTActive learning aims to reduce the labeling costs by selecting only samples thatare informative to improve the accuracy of the network. Few existing works haveaddressed the problem of active learning for object detection, and most of themestimate the informativeness of an image based only on the classification ... | wJxsc7Vk4J6 | This paper describes an mixture density estimation based method for active learning of object detection. | 5: Marginally below acceptance threshold | This paper describes a mixture density estimation based method for active learning of object detection.
pros:
1. the approach of using GMM for modeling uncertainties of object detection network prediction is new.
2. the paper has conducted extensive experimental evaluations to validate the method.
cons:
1. One key ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deep Active Learning for Object Detection with Mixture Density Networks
### Paper Abstract
Active learning aims to reduce the labeling costs by selecting only samples th... |
m2ZxDprKYlO | ICLR.cc/2021/Conference | 2021 | Meta-Learning with Implicit Processes | ["YIZHOU CHEN", "DONG LI", "NA LI", "TONG LIANG", "SHIZHUO ZHANG", "Bryan Kian Hsiang Low"] | This paper presents a novel implicit process-based meta-learning (IPML) algorithm that, in contrast to existing works, explicitly represents each task as a continuous latent vector and models its probabilistic belief within the highly expressive IP framework. Unfortunately, meta-training in IPML is computationally chal... | ["ipml", "implicit processes", "algorithm", "works", "performance", "novel implicit", "contrast", "task", "continuous latent vector", "models"] | ABSTRACTThis paper presents a novel implicit process-based meta-learning (IPML) algo-rithm that, in contrast to existing works, explicitly represents each task as a contin-uous latent vector and models its probabilistic belief within the highly expressiveIP framework. Unfortunately, meta-training in IPML is computation... | cq9gdX2FSne | Solid paper - Though, better comparison to Neural Processes is needed | 6: Marginally above acceptance threshold | === Summary ===
The paper proposes a meta-learning method based on implicit process (IP) framework in which each task is represented by a latent vector. The IP setup for meta-learning seems identical to that of Neural processes [1]. In that, the key challenge for adaptation to a task based on a context/support set $(X... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Meta-Learning with Implicit Processes
### Paper Abstract
This paper presents a novel implicit process-based meta-learning (IPML) algorithm that, in contrast to existing ... |
GMgHyUPrXa | ICLR.cc/2021/Conference | 2021 | A Design Space Study for LISTA and Beyond | ["Tianjian Meng", "Xiaohan Chen", "Yifan Jiang", "Zhangyang Wang"] | In recent years, great success has been witnessed in building problem-specific deep networks from unrolling iterative algorithms, for solving inverse problems and beyond. Unrolling is believed to incorporate the model-based prior with the learning capacity of deep learning. This paper revisits \textit{the role of unrol... | ["lista", "design space study", "deep networks", "better", "searched top architectures", "recent years", "great success", "iterative algorithms", "inverse problems", "unrolling"] | ABSTRACTIn recent years, great success has been witnessed in building problem-specificdeep networks from unrolling iterative algorithms, for solving inverse problemsand beyond. Unrolling is believed to incorporate the model-based prior with thelearning capacity of deep learning. This paper revisits the role of unrollin... | oZILNiRwWcc | Contains extensive experiments but technical contributions are limited | 6: Marginally above acceptance threshold | This paper conducts an empirical study on unrolling architecture design, by applying NAS to search the connectivity pattern based on LISTA, and compares the searched model to the original LISTA.
Pros:
+ The paper is clearly written and not hard to follow
+ Experiments show that the searched architecture performs bette... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Design Space Study for LISTA and Beyond
### Paper Abstract
In recent years, great success has been witnessed in building problem-specific deep networks from unrolling ... |
BJlpCsC5Km | ICLR.cc/2019/Conference | 2019 | Learning Gibbs-regularized GANs with variational discriminator reparameterization | ["Nicholas Rhinehart", "Anqi Liu", "Kihyuk Sohn", "Paul Vernaza"] | We propose a novel approach to regularizing generative adversarial networks (GANs) leveraging learned {\em structured Gibbs distributions}. Our method consists of reparameterizing the discriminator to be an explicit function of two densities: the generator PDF $q$ and a structured Gibbs distribution $\nu$. Leveragin... | ["deep generative models", "graphical models", "trajectory forecasting", "GANs", "density estimation", "structured prediction"] | ABSTRACTWe propose a novel approach to regularizing generative adversarial networks(GANs) leveraging learned structured Gibbs distributions . Our method consists ofreparameterizing the discriminator to be an explicit function of two densities: thegenerator PDF qand a structured Gibbs distribution . Leveraging recent wo... | HklfE4KKh7 | Decent idea but need more motivating experiments. | 5: Marginally below acceptance threshold | The paper proposes to reparameterize the discriminator to be an explicit function of two densities so that one could inject domain specific knowledge easily. As the authors say, that one way to inject domain specific information is by learning an energy function. Making use of this intuition, authors proposed to regul... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Gibbs-regularized GANs with variational discriminator reparameterization
### Paper Abstract
We propose a novel approach to regularizing generative adversarial ... |
RqN8W3R76J | logconference.io/LOG/2022/Conference | 2022 | AutoGDA: Automated Graph Data Augmentation for Node Classification | ["Tong Zhao", "Xianfeng Tang", "Danqing Zhang", "Haoming Jiang", "Nikhil Rao", "Yiwei Song", "Pallav Agrawal", "Karthik Subbian", "Bing Yin", "Meng Jiang"] | Graph data augmentation has been used to improve generalizability of graph machine learning. However, by only applying fixed augmentation operations on entire graphs, existing methods overlook the unique characteristics of communities which naturally exist in the graphs. For example, different communities can have vari... | ["Graph Neural Networks", "Data Augmentation"] | AutoGDA: Automated Graph Data Augmentationfor Node ClassificationTong Zhao ♥, Xianfeng Tang|, Danqing Zhang|, Haoming Jiang|, Nikhil Rao|,Yiwei Song|, Pallav Agrawal|, Karthik Subbian|, Bing Yin|, Meng JiangUniversity of Notre Dame, Notre Dame, IN, USA♥Snap Inc., Seattle, WA, USA|Amazon.com Inc., Palo Alto, CA, USAAbst... | u8h5q4IQ3k | This paper studies the graph structure learning (GSL) problem, and proposes a bilevel optimization framework AutoGDA. Despite the effectiveness of some small datasets, I would recommend rejection due to the poor discussion and comparison of the studied area.
Strong points
S1. The proposed method handles an important ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
AutoGDA: Automated Graph Data Augmentation for Node Classification
### Paper Abstract
Graph data augmentation has been used to improve generalizability of graph machine ... | |||
g6OrH2oT5so | ICLR.cc/2021/Conference | 2021 | Bridging the Imitation Gap by Adaptive Insubordination | ["Luca Weihs", "Unnat Jain", "Jordi Salvador", "Svetlana Lazebnik", "Aniruddha Kembhavi", "Alex Schwing"] | When expert supervision is available, practitioners often use imitation learning with varying degrees of success. We show that when an expert has access to privileged information that is unavailable to the student, this information is marginalized in the student policy during imitation learning resulting in an ''imitat... | ["Privileged Experts", "Imitation Learning", "Reinforcement Learning", "Actor-Critic", "Behavior Cloning", "MiniGrid", "Knowledge Distillation"] | ABSTRACTWhen expert supervision is available, practitioners often use imitation learning withvarying degrees of success. We show that when an expert has access to privilegedinformation that is unavailable to the student, this information is marginalizedin the student policy during imitation learning resulting in an “im... | Z8575NVVuj | Introducing a critical issue in imitation learning with a simple solution | 6: Marginally above acceptance threshold | ### Summary
This paper identifies a problem in imitation learning when an expert has access to privileged information that is not available to the learner. When a decision has to be made based on the privileged information, the learner tends to choose average or uniformly random actions of the expert due to the lack of... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Bridging the Imitation Gap by Adaptive Insubordination
### Paper Abstract
When expert supervision is available, practitioners often use imitation learning with varying d... |
5RqtEQc4izA | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | Time-Varying Word Clouds | ["Anonymous"] | We visualize time-varying text information with physically based simulation. Word-clouds are a popular means of visualizing the frequency of different concepts in a text document, but there is little work using text that has a time component, for instance, news feeds, twitter, or abstracts of papers published in a giv... | ["word cloud", "time-varying", "real time", "dynamic"] | ABSTRACTWe visualize time-varying text information with physically basedsimulation. Word-clouds are a popular means of visualizing thefrequency of different concepts in a text document, but there is littlework using text that has a time component, for instance, news feeds,twitter, or abstracts of papers published in a ... | -Bf1JLaqjbp | So fun! | 8: Top 50% of accepted papers, clear accept | When I saw the video for this paper, my reaction was "this is so fun! I wish I had done this!"
I feel like the *concept* is in itself an extremely strong part of this submission. Independent of implementation details, I feel like the concept itself is an exciting point of merit for publication of this submission.
It'... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Time-Varying Word Clouds
### Paper Abstract
We visualize time-varying text information with physically based simulation. Word-clouds are a popular means of visualizing ... |
HJTzHtqee | ICLR.cc/2017/conference | 2017 | A Compare-Aggregate Model for Matching Text Sequences | ["Shuohang Wang", "Jing Jiang"] | Many NLP tasks including machine comprehension, answer selection and text entailment require the comparison between sequences. Matching the important units between sequences is a key to solve these problems. In this paper, we present a general "compare-aggregate" framework that performs word-level matching followed by ... | ["Natural language processing", "Deep learning"] | ABSTRACTMany NLP tasks including machine comprehension, answer selection and text en-tailment require the comparison between sequences. Matching the important unitsbetween sequences is a key to solve these problems. In this paper, we present ageneral “compare-aggregate” framework that performs word-level matching fol-l... | H1th_uZNg | A solid empirical study | 7: Good paper, accept | This paper proposes a compare-aggregate framework that performs word-level matching followed by aggregation with convolutional neural networks. It compares six different comparison functions and evaluates them on four datasets. Extensive experimental results have been reported and compared against various published bas... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Compare-Aggregate Model for Matching Text Sequences
### Paper Abstract
Many NLP tasks including machine comprehension, answer selection and text entailment require the... |
Hke20iA9Y7 | ICLR.cc/2019/Conference | 2019 | Efficient Training on Very Large Corpora via Gramian Estimation | ["Walid Krichene", "Nicolas Mayoraz", "Steffen Rendle", "Li Zhang", "Xinyang Yi", "Lichan Hong", "Ed Chi", "John Anderson"] | We study the problem of learning similarity functions over very large corpora using neural network embedding models. These models are typically trained using SGD with random sampling of unobserved pairs, with a sample size that grows quadratically with the corpus size, making it expensive to scale.
We propose new effic... | ["similarity learning", "pairwise learning", "matrix factorization", "Gramian estimation", "variance reduction", "neural embedding models", "recommender systems"] | ABSTRACTWe study the problem of learning similarity functions over very large corporausing neural network embedding models. These models are typically trained usingSGD with random sampling of unobserved pairs, with a sample size that growsquadratically with the corpus size, making it expensive to scale. We propose newe... | rJgsqcLt3m | Nice work | 8: Top 50% of accepted papers, clear accept | This paper proposes an efficient algorithm to learn neural embedding models with a dot-product structure over very large corpora. The main method is to reformulate the objective function in terms of generalized Gramiam matrices, and maintain estimates of those matrices in the training process. The algorithm uses less ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Efficient Training on Very Large Corpora via Gramian Estimation
### Paper Abstract
We study the problem of learning similarity functions over very large corpora using ne... |
BVSM0x3EDK6 | ICLR.cc/2021/Conference | 2021 | Robust and Generalizable Visual Representation Learning via Random Convolutions | ["Zhenlin Xu", "Deyi Liu", "Junlin Yang", "Colin Raffel", "Marc Niethammer"] | While successful for various computer vision tasks, deep neural networks have shown to be vulnerable to texture style shifts and small perturbations to which humans are robust. In this work, we show that the robustness of neural networks can be greatly improved through the use of random convolutions as data augmentatio... | ["domain generalization", "robustness", "representation learning", "data augmentation"] | ABSTRACTWhile successful for various computer vision tasks, deep neural networks haveshown to be vulnerable to texture style shifts and small perturbations to whichhumans are robust. In this work, we show that the robustness of neural networks canbe greatly improved through the use of random convolutions as data augmen... | 7GQvGH93Mtw | Interesting Method, Several Issues in Experiments | 6: Marginally above acceptance threshold | Summary: This paper uses random convolution augmentation to learn representation that is robust for transferring across domains.
Quality: The paper is well-written, easy to following. The method is presented clearly, and the enough experimental evaluation was conducted. Overall, the quality is good.
Clarity: The meth... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust and Generalizable Visual Representation Learning via Random Convolutions
### Paper Abstract
While successful for various computer vision tasks, deep neural networ... |
BJlBSkHtDS | ICLR.cc/2020/Conference | 2020 | Padé Activation Units: End-to-end Learning of Flexible Activation Functions in Deep Networks | ["Alejandro Molina", "Patrick Schramowski", "Kristian Kersting"] | The performance of deep network learning strongly depends on the choice of the non-linear activation function associated with each neuron. However, deciding on the best activation is non-trivial and the choice depends on the architecture, hyper-parameters, and even on the dataset. Typically these activations are fixed ... | ["paus", "pad\u00e9 activation units", "learning", "flexible activation functions", "deep networks", "choice", "performance", "deep network", "activation function"] | ABSTRACTThe performance of deep network learning strongly depends on the choice of thenon-linear activation function associated with each neuron. However, decidingon the best activation is non-trivial, and the choice depends on the architecture,hyper-parameters, and even on the dataset. Typically these activations are ... | HJgsBTOLKr | Official Blind Review #2 | 6: Weak Accept | This work proposes an activation function that contain parameters to be learned through training. The idea is to give the learning algorithm more "freedom" to choose a good activation function, and hopefully better performance can be achieved.
The paper is well written, and the experiment results look reasonable. Howe... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Padé Activation Units: End-to-end Learning of Flexible Activation Functions in Deep Networks
### Paper Abstract
The performance of deep network learning strongly depends... | |
SkxEWgStDr | ICLR.cc/2020/Conference | 2020 | A Simple Geometric Proof for the Benefit of Depth in ReLU Networks | ["Asaf Amrami", "Yoav Goldberg"] | We present a simple proof for the benefit of depth in multi-layer feedforward network with rectifed activation (``"depth separation"). Specifically we present a sequence of classification problems f_i such that (a) for any fixed depth rectified network we can find an index m such that problems with index > m require ex... | ["benefit", "depth", "simple geometric proof", "relu networks", "index", "simple proof", "feedforward network", "rectifed activation", "depth separation", "sequence"] | ABSTRACTWe present a simple proof for the benefit of depth in multi-layer feedforwardnetwork with rectifed activation (“depth separation”). Specifically we present asequence of classification problems fisuch that (a) for any fixed depth rectifiednetwork we can find an index msuch that problems with index >m require exp... | SJlj-VnNtr | Official Blind Review #2 | 3: Weak Reject | This paper presents a proof that deeper networks need less units than shallower ones for a family of problems.
More exactly, the authors exhibit a series of indexed problems f_i, such that:
- for a given depth, there are problems that require a exponential number of units to be approximated by a network, and
- for any... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Simple Geometric Proof for the Benefit of Depth in ReLU Networks
### Paper Abstract
We present a simple proof for the benefit of depth in multi-layer feedforward netwo... | |
HJeu43ActQ | ICLR.cc/2019/Conference | 2019 | NOODL: Provable Online Dictionary Learning and Sparse Coding | ["Sirisha Rambhatla", "Xingguo Li", "Jarvis Haupt"] | We consider the dictionary learning problem, where the aim is to model the given data as a linear combination of a few columns of a matrix known as a dictionary, where the sparse weights forming the linear combination are known as coefficients. Since the dictionary and coefficients, parameterizing the linear model are ... | ["dictionary learning", "provable dictionary learning", "online dictionary learning", "sparse coding", "support recovery", "iterative hard thresholding", "matrix factorization", "neural architectures", "neural networks", "noodl"] | ABSTRACTWe consider the dictionary learning problem, where the aim is to model the givendata as a linear combination of a few columns of a matrix known as a dictionary ,where the sparse weights forming the linear combination are known as coeffi-cients . Since the dictionary and coefficients, parameterizing the linear m... | B1xtSy6t2Q | Interesting paper, clear contribution. More intuition would be great. | 6: Marginally above acceptance threshold | The paper deals with the problem of recovering an exact solution for both the dictionary and the activation coefficients. As other works, the solution is based on a proper initialization of the dictionary. The authors suggest using Aurora 2015 as a possible initialization. The contribution improves Arora 2015 in that i... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NOODL: Provable Online Dictionary Learning and Sparse Coding
### Paper Abstract
We consider the dictionary learning problem, where the aim is to model the given data as ... |
vY0bnzBBvtr | ICLR.cc/2021/Conference | 2021 | Provably More Efficient Q-Learning in the One-Sided-Feedback/Full-Feedback Settings | ["Xiao-Yue Gong", "David Simchi-Levi"] | Motivated by the episodic version of the classical inventory control problem, we propose a new Q-learning-based algorithm, Elimination-Based Half-Q-Learning (HQL), that enjoys improved efficiency over existing algorithms for a wide variety of problems in the one-sided-feedback setting. We also provide a simpler variant... | ["Q-learning", "episodic MDP", "full-feedback", "one-sided-feedback", "inventory control", "inventory"] | ABSTRACTMotivated by the episodic version of the classical inventory control problem, wepropose a new Q-learning-based algorithm, Elimination-Based Half-Q-Learning(HQL) , that enjoys improved efficiency over existing algorithms for a wide varietyof problems in the one-sided-feedback setting. We also provide a simpler v... | pCrX_TCzYsL | Review for "Provably More Efficient Q-Learning in the One-Sided-Feedback/Full-Feedback Settings" | 4: Ok but not good enough - rejection | This paper proposes Q-learning based algorithms called Elimination-Based Half-Q-Learning (HQL) and Full-Q-Learning (FQL). In the one-sided-feedback setting, the proposed algorithm improves the regret bounds over existing methods in terms of the dependency on the size of state-action space. Numerical experiments are pro... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Provably More Efficient Q-Learning in the One-Sided-Feedback/Full-Feedback Settings
### Paper Abstract
Motivated by the episodic version of the classical inventory contr... |
K4wkUp5xNK | ICLR.cc/2021/Conference | 2021 | Invariant Causal Representation Learning | ["Chaochao Lu", "Yuhuai Wu", "Jos\u00e9 Miguel Hern\u00e1ndez-Lobato", "Bernhard Sch\u00f6lkopf"] | Due to spurious correlations, machine learning systems often fail to generalize to environments whose distributions differ from the ones used at training time. Prior work addressing this, either explicitly or implicitly, attempted to find a data representation that has an invariant causal relationship with the outcome.... | ["outcome", "invariant causal representation", "environments", "data representation", "generalization guarantees", "nonlinear setting", "due", "spurious correlations", "machine", "systems"] | ABSTRACTDue to spurious correlations, machine learning systems often fail to generalize toenvironments whose distributions differ from the ones used at training time. Priorwork addressing this, either explicitly or implicitly, attempted to find a data repre-sentation that has an invariant causal relationship with the o... | prtT9EC0R9L | Interesting topic but the assumption and the experiments are not convincing. | 4: Ok but not good enough - rejection | This paper proposes an invariant causal representation learning paradigm in the nonlinear setting. Based on a conditional factorial assumption, they proved identifiability up to a linear transform. The ICRL objective, in this case, is able to discover all the direct causes of the outcome, and thus enables OOD generaliz... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Invariant Causal Representation Learning
### Paper Abstract
Due to spurious correlations, machine learning systems often fail to generalize to environments whose distrib... |
PEcNk5Bad7z | ICLR.cc/2021/Conference | 2021 | Learning Irreducible Representations of Noncommutative Lie Groups | ["Noah Shutty", "Casimir Wierzynski"] | Recent work has constructed neural networks that are equivariant to continuous symmetry groups such as 2D and 3D rotations. This is accomplished using explicit group representations to derive the equivariant kernels and nonlinearities. We present two contributions motivated by frontier applications of equivariance beyo... | ["equivariance", "object tracking", "equivariant neural networks", "deep learning", "point cloud", "lie group", "lie algebra", "lorentz group", "poincar\u00e9 group"] | ABSTRACTRecent work has constructed neural networks that are equivariant to continuoussymmetry groups such as 2D and 3D rotations. This is accomplished using explicitgroup representations to derive the equivariant kernels and nonlinearities. Wepresent two contributions motivated by frontier applications of equivariance... | 8lMNXKjPPr | Learning Irreducible Representations of Noncommutative Lie Groups review | 5: Marginally below acceptance threshold | Paper summary:
The paper proposes the algorithm LearnRep that uses gradient descent methods to learn Lie algebras from structure constants, before obtaining the corresponding group representation through the exponential map. The algorithm is tested on SO(3), SO(2, 1), and SO(3, 1). In addition to this, the paper propo... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Irreducible Representations of Noncommutative Lie Groups
### Paper Abstract
Recent work has constructed neural networks that are equivariant to continuous symme... |
YdFV691VRcg | ICLR.cc/2021/Conference | 2021 | Learning Semantic Similarities for Prototypical Classifiers | ["Joao Monteiro", "Isabela Albuquerque", "Jahangir Alam", "Tiago Falk"] | Recent metric learning approaches parametrize semantic similarity measures through the use of an encoder trained along with a similarity model, which operates over pairs of representations. We extend such a setting and enable its use in tasks including multi-class classification in order to tackle known issues observed... | ["Semantic similarities", "metric learning", "prototypical classifiers", "adversarial robustness"] | ABSTRACTRecent metric learning approaches parametrize semantic similarity measuresthrough the use of an encoder trained along with a similarity model, which op-erates over pairs of representations. We extend such a setting and enable its use intasks including multi-class classification in order to tackle known issues o... | NhtUA0JRdbE | An interesting approach that has not been elaborated and studied carefully enough. | 4: Ok but not good enough - rejection | In order to organize my review, I use the NeurIPS2020 template with slight adaptions to fit the ICLR requirements.
### Summary and contributions: Briefly summarize the paper and its contributions.
The paper proposes a method to learn an encoder, a similarity measure, and a set of prototypes in an end-to-end setting.... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Semantic Similarities for Prototypical Classifiers
### Paper Abstract
Recent metric learning approaches parametrize semantic similarity measures through the use... |
HJDBUF5le | ICLR.cc/2017/conference | 2017 | Towards a Neural Statistician | ["Harrison Edwards", "Amos Storkey"] | An efficient learner is one who reuses what they already know to tackle a new problem. For a machine learner, this means understanding the similarities amongst datasets. In order to do this, one must take seriously the idea of working with datasets, rather than datapoints, as the key objects to model. Towards this goal... | ["datasets", "statistics", "towards", "neural statistician", "network", "new datasets", "neural statistician towards", "efficient learner", "new problem", "machine learner"] | ABSTRACTAnefficient learner is one who reuses what they already know to tackle a newproblem. For a machine learner, this means understanding the similarities amongstdatasets. In order to do this, one must take seriously the idea of working withdatasets, rather than datapoints, as the key objects to model. Towards this ... | Sy8k9FbNg | Interesting paper that starts to expand the repertoire of variational autoencoders | 6: Marginally above acceptance threshold | The authors introduce a variant of the variational autoencoder (VAE) that models dataset-level latent variables. The idea is clearly motivated and well described. In my mind the greatest contribution of this paper is the movement beyond the relatively simple graphical model structure of the traditional VAEs and the int... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards a Neural Statistician
### Paper Abstract
An efficient learner is one who reuses what they already know to tackle a new problem. For a machine learner, this means... |
VH6WIPF4Sj | robot-learning.org/CoRL/2023/Conference | 2023 | Predicting Object Interactions with Behavior Primitives: An Application in Stowing Tasks | ["Haonan Chen", "Yilong Niu", "Kaiwen Hong", "Shuijing Liu", "Yixuan Wang", "Yunzhu Li", "Katherine Rose Driggs-Campbell"] | Stowing, the task of placing objects in cluttered shelves or bins, is a common task in warehouse and manufacturing operations. However, this task is still predominantly carried out by human workers as stowing is challenging to automate due to the complex multi-object interactions and long-horizon nature of the task. Pr... | ["Robotic Manipulation", "Model Learning", "Graph-Based Neural Dynamics", "Multi-Object Interactions"] | Predicting Object Interactions with BehaviorPrimitives: An Application in Stowing TasksHaonan Chen, Yilong Niu∗ ∗, Kaiwen Hong∗, Shuijing Liu, Yixuan Wang,Yunzhu Li ,Katherine Driggs-CampbellUniversity of Illinois, Urbana-Champaign{haonan2, yilongn2, kaiwen2, sliu105, yixuan22, yunzhuli, krdc }@illinois.eduAbstract: St... | HF_BiEQVWhd | Strengths of the paper:
1: The stowing task is novel and important for the robot manipulation community.
2: This approach is trained purely in simulation and can generalize to real-world scenarios.
3: The real-robot demos look cool and interesting.
Weakness of the paper:
1: the author proposes a graph neural net... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Predicting Object Interactions with Behavior Primitives: An Application in Stowing Tasks
### Paper Abstract
Stowing, the task of placing objects in cluttered shelves or ... | ||
HJIoJWZCZ | ICLR.cc/2018/Conference | 2018 | Adversarial Dropout Regularization | ["Kuniaki Saito", "Yoshitaka Ushiku", "Tatsuya Harada", "Kate Saenko"] | We present a domain adaptation method for transferring neural representations from label-rich source domains to unlabeled target domains. Recent adversarial methods proposed for this task learn to align features across domains by ``fooling'' a special domain classifier network. However, a drawback of this approach is t... | ["domain adaptation", "computer vision", "generative models"] | ABSTRACTWe present a domain adaptation method for transferring neural representationsfrom label-rich source domains to unlabeled target domains. Recent adversarialmethods proposed for this task learn to align features across domains by “fool-ing” a special domain classifier network. However, a drawback of this approach... | HJ4p6dFeG | Review | 5: Marginally below acceptance threshold | (Summary)
This paper is about learning discriminative features for the target domain in unsupervised DA problem. The key idea is to use a critic which randomly drops the activations in the logit and maximizes the sensitivity between two versions of discriminators.
(Pros)
The approach proposed in section 3.2 uses dropo... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adversarial Dropout Regularization
### Paper Abstract
We present a domain adaptation method for transferring neural representations from label-rich source domains to unl... |
FRKBdXhkQE0 | robot-learning.org/CoRL/2023/Conference | 2023 | FastRLAP: A System for Learning High-Speed Driving via Deep RL and Autonomous Practicing | ["Kyle Stachowicz", "Dhruv Shah", "Arjun Bhorkar", "Ilya Kostrikov", "Sergey Levine"] | We present a system that enables an autonomous small-scale RC car to drive aggressively from visual observations using reinforcement learning (RL). Our system, FastRLAP, trains autonomously in the real world, without human interventions, and without requiring any simulation or expert demonstrations. Our system integrat... | ["reinforcement learning", "offroad driving", "vision-based navigation"] | FastRLAP: A System for Learning High-SpeedDriving via Deep RL and Autonomous PracticingKyle Stachowicz†, Dhruv Shah†, Arjun Bhorkar†, Ilya Kostrikov, Sergey LevineUC BerkeleyAbstract: We present a system that enables a 1/10th-scale autonomous car to driveat high speeds from visual observations using reinforcement learn... | TnaeaUc1KTr | The paper is well written and easy to follow.
# Strengths
- The use of an FSM to provide intermediate checkpoints and a collision-recovery subroutine to facilitate “in the wild” reinforcement learning in a sample-efficient, reset-free manner is impressive, novel as far as I’m aware, and brings the field closer to mak... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
FastRLAP: A System for Learning High-Speed Driving via Deep RL and Autonomous Practicing
### Paper Abstract
We present a system that enables an autonomous small-scale RC... | ||
HygS91rYvH | ICLR.cc/2020/Conference | 2020 | Universal Adversarial Attack Using Very Few Test Examples | ["Amit Deshpande", "Sandesh Kamath", "K V Subrahmanyam"] | Adversarial attacks such as Gradient-based attacks, Fast Gradient Sign Method (FGSM) by Goodfellow et al.(2015) and DeepFool by Moosavi-Dezfooli et al. (2016) are input-dependent, small pixel-wise perturbations of images which fool state of the art neural networks into misclassifying images but are unlikely to fool any... | ["universal", "adversarial", "SVD"] | ABSTRACTAdversarial attacks such as Gradient-based attacks, Fast Gradient Sign Method(FGSM ) by Goodfellow et al. (2015) and DeepFool by Moosavi-Dezfooli et al.(2016) are input-dependent, small pixel-wise perturbations of images which foolstate of the art neural networks into misclassifying images but are unlikely to f... | HJxMVJT3KS | Official Blind Review #1 | 3: Weak Reject | This paper studied the problem of universal adversarial attack which is an input-agnostic perturbation. The authors proposed to use the top singular vector of input-dependent adversarial attack directions to perform universal adversarial attacks. The authors evaluated the error rates and fooling rates for three attacks... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Universal Adversarial Attack Using Very Few Test Examples
### Paper Abstract
Adversarial attacks such as Gradient-based attacks, Fast Gradient Sign Method (FGSM) by Good... | |
H1xsSjC9Ym | ICLR.cc/2019/Conference | 2019 | Learning to Understand Goal Specifications by Modelling Reward | ["Dzmitry Bahdanau", "Felix Hill", "Jan Leike", "Edward Hughes", "Arian Hosseini", "Pushmeet Kohli", "Edward Grefenstette"] | Recent work has shown that deep reinforcement-learning agents can learn to follow language-like instructions from infrequent environment rewards. However, this places on environment designers the onus of designing language-conditional reward functions which may not be easily or tractably implemented as the complexity o... | ["instruction following", "reward modelling", "language understanding"] | ABSTRACTRecent work has shown that deep reinforcement-learning agents can learn to followlanguage-like instructions from infrequent environment rewards. However, thisplaces on environment designers the onus of designing language-conditional rewardfunctions which may not be easily or tractably implemented as the complex... | rklL8IgSn7 | This is a well-written paper. The idea of learning rewards from instructions is interesting although quite straightforward. The experiments show improvement over baseline reward, but they were performed on simple synthetic tasks. | 6: Marginally above acceptance threshold | The paper presents an approach for simultaneously learning policies and reward functions for reaching goals that are described by an instruction providing spatial relations among objects. The proposed platform, called Adversarial Goal-Induced Learning from Examples (AGILE), is composed of an off-the-shelf RL module lik... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Understand Goal Specifications by Modelling Reward
### Paper Abstract
Recent work has shown that deep reinforcement-learning agents can learn to follow langu... |
rylKB3A9Fm | ICLR.cc/2019/Conference | 2019 | Assessing Generalization in Deep Reinforcement Learning | ["Charles Packer*", "Katelyn Gao*", "Jernej Kos", "Philipp Krahenbuhl", "Vladlen Koltun", "Dawn Song"] | Deep reinforcement learning (RL) has achieved breakthrough results on many tasks, but has been shown to be sensitive to system changes at test time. As a result, building deep RL agents that generalize has become an active research area. Our aim is to catalyze and streamline community-wide progress on this problem by p... | ["reinforcement learning", "generalization", "benchmark"] | ABSTRACTDeep reinforcement learning (RL) has achieved breakthrough results on manytasks, but has been shown to be sensitive to system changes at test time. As aresult, building deep RL agents that generalize has become an active researcharea. Our aim is to catalyze and streamline community-wide progress on thisproblem ... | B1lNZjECnQ | Review | 5: Marginally below acceptance threshold | This paper proposes a benchmark for for reinforcement learning to study generalization in stationary and changing environments. A combination of several existing env. from OpenAi gym is taken and several ways to set this parameters is proposed. Paper provides a relatively thorough study of popular methodologies on this... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Assessing Generalization in Deep Reinforcement Learning
### Paper Abstract
Deep reinforcement learning (RL) has achieved breakthrough results on many tasks, but has been... |
8nXkyH2_s6 | ICLR.cc/2021/Conference | 2021 | Neural networks behave as hash encoders: An empirical study | ["Fengxiang He", "Shiye Lei", "Jianmin Ji", "Dacheng Tao"] | The input space of a neural network with ReLU-like activations is partitioned into multiple linear regions, each corresponding to a specific activation pattern of the included ReLU-like activations. We demonstrate that this partition exhibits the following encoding properties across a variety of deep learning models: (... | ["Explainability of deep learning"] | ABSTRACTThe input space of a neural network with ReLU-like activations is partitioned intomultiple linear regions, each corresponding to a specific activation pattern of theincluded ReLU-like activations. We demonstrate that this partition exhibits thefollowing encoding properties across a variety of deep learning mode... | bIOWQHLFaJN | MNIST alone is not enough to draw any decisive conclusions | 5: Marginally below acceptance threshold | This draft proposes to use the relu activation pattern of the neurons in the neural network as the hash code for the input. Essentially the input features are bucketized into small piecewise linear regions. The authors show empirically that the proposed hash code has small collision and high accuracy given certain cond... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural networks behave as hash encoders: An empirical study
### Paper Abstract
The input space of a neural network with ReLU-like activations is partitioned into multipl... |
rylWVnR5YQ | ICLR.cc/2019/Conference | 2019 | Context Dependent Modulation of Activation Function | ["Long Sha", "Jonathan Schwarcz", "Pengyu Hong"] | We propose a modification to traditional Artificial Neural Networks (ANNs), which provides the ANNs with new aptitudes motivated by biological neurons. Biological neurons work far beyond linearly summing up synaptic inputs and then transforming the integrated information. A biological neuron change firing modes accor... | ["Artificial Neural Network", "Convolution Neural Network", "Long Short-Term Memory", "Activation Function", "Neuromodulation"] | ABSTRACTWe propose a modification to traditional Artificial Neural Networks (ANNs),which provides the ANNs with new aptitudes motivated by biological neurons.Biological neurons work far beyond linearly summing up synaptic inputs and thentransforming the integrated information. A biological neuron change firing modesacc... | HJgd_ZWanX | Interesting, but no convincing results and analysis | 4: Ok but not good enough - rejection | This paper proposes a scalar modulator adding to hidden nodes before an activation function. The authors claim that it controls the sensitivity of the hidden nodes by changing the slope of activation function. The modulator is combined with a simple CNN, DesneNet, and a LSTM model, and they provided the performance imp... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Context Dependent Modulation of Activation Function
### Paper Abstract
We propose a modification to traditional Artificial Neural Networks (ANNs), which provides the ANN... |
PUQAhZ2jTr | eswc-conferences.org/ESWC/2021/Conference/Industry_Track | 2021 | Maintain electrical power system model quality with Shape Constraints | ["Merlin B\u00f6gershausen", "Oliver Scheufeld", "Christoph Lange", "Oya Beyan"] | The models of the interconnected European power system is exchanged in an RDF based format.
To ensure stability, a performant and repeatable evaluation of the data quality is crucial.
We will show how evacuations are possible with semantic web technologies. | ["Data Quality", "Electrical Power Engineering", "Power System Modelling", "SHACL"] | Maintain electrical power system model qualitywith Shape ConstraintsMerlin Bögershausen1,3, Oliver Scheufeld1, ChristophLange2,3[0000−0001−9879−3827], and Oya Deniz Beyan2,3[0000−0001−7611−3501]1SOPTIM AG <first>.<last>@soptim.de2RWTH Aachen <first>.<last>@rwth-aachen.de3Fraunhofer Institute for Applied Information Te... | zpZeWKEv0e | Review: Maintain electrical power system model quality with Shape Constraints | 5: Marginally below acceptance threshold | In their article, the authors demonstrate how SHACL can be utilized to maintain the quality of ENTSO-E Common Grid Model Exchange Specification (CGMES) data.
My main question regarding this contribution is how it compares to
- "Nenadić, Kosa R., Milan M. Gavrić, and Vladimir I. Đurđević. "Validation of CIM datasets... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Maintain electrical power system model quality with Shape Constraints
### Paper Abstract
The models of the interconnected European power system is exchanged in an RDF ba... |
w5bNwUzj33 | ICLR.cc/2021/Conference | 2021 | Cross-Domain Few-Shot Learning by Representation Fusion | ["Thomas Adler", "Johannes Brandstetter", "Michael Widrich", "Andreas Mayr", "David Kreil", "Michael K Kopp", "G\u00fcnter Klambauer", "Sepp Hochreiter"] | In order to quickly adapt to new data, few-shot learning aims at learning from few examples, often by using already acquired knowledge. The new data often differs from the previously seen data due to a domain shift, that is, a change of the input-target distribution. While several methods perform well on small domain s... | ["cross-domain learning", "few-shot learning", "Hebbian learning", "ensemble learning", "domain shift", "domain adaptation", "representation fusion"] | ABSTRACTIn order to quickly adapt to new data, few-shot learning aims at learning from fewexamples, often by using already acquired knowledge. The new data often differsfrom the previously seen data due to a domain shift, that is, a change of the input-target distribution. While several methods perform well on small do... | 7YggefBKGf | CROSS-DOMAIN FEW-SHOT LEARNING BY REPRESENTATION FUSION | 4: Ok but not good enough - rejection | Summary: This paper proposes a domain-shift problem using fewer training examples. It suggests representation fusion as the concept of unifying and merging information from different layers of abstraction.
Cross-domain Hebbian Ensemble Few-shot learning (CHEF) is introduced for extracting features using representatio... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Cross-Domain Few-Shot Learning by Representation Fusion
### Paper Abstract
In order to quickly adapt to new data, few-shot learning aims at learning from few examples, o... |
Syfe2iR5FQ | ICLR.cc/2019/Conference | 2019 | Parametrizing Fully Convolutional Nets with a Single High-Order Tensor | ["Jean Kossaifi", "Adrian Bulat", "Georgios Tzimiropoulos", "Maja Pantic"] | Recent findings indicate that over-parametrization, while crucial to the success of deep learning, also introduces large amounts of redundancy. Tensor methods have the potential to parametrize over-complete representations in a compact manner by leveraging this redundancy. In this paper, we propose fully parametrizing ... | ["single", "tensor", "number", "convolutional nets", "redundancy", "cnn", "tensor recent findings", "crucial", "success", "deep learning"] | ABSTRACTRecent findings indicate that over-parametrization, while crucial to the success ofdeep learning, also introduces large amounts of redundancy. Tensor methods havethe potential to parametrize over-complete representations in a compact manner byleveraging this redundancy. In this paper, we propose fully parametri... | rylKhzdv27 | the paper starts with a nice idea, but the development is not clear enough | 3: Clear rejection | This paper starts with a nice and appealing idea --- parametrizing a CNN using just one high-order tensor. The modes of the tensor represent different "coordinates" of the weights of CNN. This sounds very reasonable.
The claimed contributions are also appealing: by doing so, the tensor representation can be used for 1... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Parametrizing Fully Convolutional Nets with a Single High-Order Tensor
### Paper Abstract
Recent findings indicate that over-parametrization, while crucial to the succes... |
trB5knNIY61 | graphicsinterface.org/Graphics_Interface/2022/Conference | 2022 | Pointing Tasks using Spatial Audio on Smartphones for People with Vision Impairments | ["Anonymous"] | We present an experimental investigation of spatial audio feedback using smartphones to support direction localization in pointing tasks for people with visual impairments (PVIs). We do this using a mobile game based on a bow-and-arrow metaphor. Our game provides a combination of spatial and non-spatial (sound beacon)... | ["Human-centered computing - Auditory Feedback", "Computing methodologies - Mixed / Augmented reality", "Human-centered computing - Empirical studies in accessibility"] | ABSTRACTWe present an experimental investigation of spatial audio feedbackusing smartphones to support direction localization in pointing tasksfor people with visual impairments (PVIs). We do this using a mobilegame based on a bow-and-arrow metaphor. Our game provides acombination of spatial and non-spatial (sound beac... | qBQEaG5qd7 | Paper 13 review | 5: Marginally below acceptance threshold | The paper describes an experiment to investigate spatial audio feedback using smartphones for people with visual impairments. The research discusses how it can support direction localization in pointing tasks by using a mobile game with the metaphor of a bow-and-arrow task. The paper discusses the efficacy of spatial a... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Pointing Tasks using Spatial Audio on Smartphones for People with Vision Impairments
### Paper Abstract
We present an experimental investigation of spatial audio feedbac... |
jiNcjRzB46 | NeurIPS.cc/2022/Workshop/Offline_RL | 2022 | ABC: Adversarial Behavioral Cloning for Offline Mode-Seeking Imitation Learning | ["Eddy Hudson", "Ishan Durugkar", "Garrett Warnell", "Peter Stone"] | Given a dataset of interactions with an environment of interest, a viable method to extract an agent policy is to estimate the maximum likelihood policy indicated by this data. This approach is commonly referred to as behavioral cloning (BC). In this work, we describe a key disadvantage of BC that arises due to the max... | ["Imitation Learning", "Offline Reinforcement Learning"] | ABC: Adversarial Behavioral Cloning for OfflineMode-Seeking Imitation LearningEddy Hudson1Ishan Durugkar1Garrett Warnell1,2Peter Stone1,31UT Austin2Army Research Laboratory3Sony AI{eddy,ishand,pstone}@cs.utexas.edugarrett.a.warnell.civ@army.milAbstractGiven a dataset of expert agent interactions with an environment of ... | GM0jJbIX6m | Review | 3: Clear rejection | The paper described a method for behavioral cloning based on adversarial learning.
The main issue with this method is that it has already been thoroughly studies in [1].
[1] Imitation Learning as f-Divergence Minimization
L. Ke, S. Choudhury, M. Barnes, W. Sun, G. Lee and S. Srinivasa
https://arxiv.org/abs/1905.12888 | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ABC: Adversarial Behavioral Cloning for Offline Mode-Seeking Imitation Learning
### Paper Abstract
Given a dataset of interactions with an environment of interest, a via... |
HklyMhCqYQ | ICLR.cc/2019/Conference | 2019 | Super-Resolution via Conditional Implicit Maximum Likelihood Estimation | ["Ke Li*", "Shichong Peng*", "Jitendra Malik"] | Single-image super-resolution (SISR) is a canonical problem with diverse applications. Leading methods like SRGAN produce images that contain various artifacts, such as high-frequency noise, hallucinated colours and shape distortions, which adversely affect the realism of the result. In this paper, we propose an altern... | ["super-resolution"] | ABSTRACTSingle-image super-resolution (SISR) is a canonical problem with diverse appli-cations. Leading methods like SRGAN (Ledig et al., 2017) produce images thatcontain various artifacts, such as high-frequency noise, hallucinated colours andshape distortions, which adversely affect the realism of the result. In this... | SJeKkSMYs7 | A super-resolution method with with fewer visual artifacts than the SRGAN method. | 6: Marginally above acceptance threshold | Quality: The overall quality of this paper is good. It adopts a simple but novel idea on SISR and shows clear improvement against existing method (e.g., SRGAN).
Clarify: This paper is well written and easy to follow. It shows a clear motivation for adopting the implicit probabilistic model.
Originality: To the best ... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Super-Resolution via Conditional Implicit Maximum Likelihood Estimation
### Paper Abstract
Single-image super-resolution (SISR) is a canonical problem with diverse appli... |
SJUX_MWCZ | ICLR.cc/2018/Conference | 2018 | Predict Responsibly: Increasing Fairness by Learning to Defer | ["David Madras", "Toniann Pitassi", "Richard Zemel"] | When machine learning models are used for high-stakes decisions, they should predict accurately, fairly, and responsibly. To fulfill these three requirements, a model must be able to output a reject option (i.e. say "``I Don't Know") when it is not qualified to make a prediction. In this work, we propose learning to de... | ["Fairness", "IDK", "Calibration", "Automated decision-making", "Transparency", "Accountability"] | ABSTRACTWhen machine learning models are used for high-stakes decisions, they shouldpredict accurately, fairly, and responsibly. To fulfill these three requirements, amodel must be able to output a reject option (i.e. say “I Don’t Know”) when itis not qualified to make a prediction. In this work, we propose learning to... | HJkk4w6lM | Conceptually very interesting, but lacks technical novelty | 6: Marginally above acceptance threshold | Strengths:
1. This paper proposes a novel framework for ensuring fairness in the classification pipeline. To this end, this work explores models that learn to defer.
2. The work is conceptually very interesting. The idea of learning to defer (as proposed in the paper) as a means to fairness is not only novel but also... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Predict Responsibly: Increasing Fairness by Learning to Defer
### Paper Abstract
When machine learning models are used for high-stakes decisions, they should predict acc... |
rkZOp9joz | MIDL.amsterdam/2018/Conference | 2018 | Whole Slide Image Classification of Gastric Cancer using Convolutional Neural Networks | ["Junni Shou", "Yan Li", "Guanzhen Yu", "Guannan Li"] | Gastric cancer is one of the main causes of cancer and cancer-related mortality worldwide, and the diagnosis based on histopathology images is a gold standard for gastric cancer detection. However, manual diagnosis is labor-intensive and low in inter-observer agreement. Computer-aided image analysis method were thus de... | ["gastric cancer", "main causes", "cancer", "mortality worldwide", "diagnosis", "histopathology images", "gold standard"] | Whole Slide Image Classification of Gastric Cancerusing Convolutional Neural NetworksJunni Shou1, Yan Li1, Guanzhen Yu2;, and Guannan Li1;1Awakens Intelligence Technology Co., Ltd, China2Department of Oncology, Longhua Hospital Affiliated to Shanghai Universityof Traditional Chinese Medicine, Shanghai, Chinaqiaoshanqia... | SkQK51b0f | Whole Slide Image Classification of Gastric Cancer using Convolutional Neural Networks | 2: Marginally below acceptance threshold | In this paper, whole-slide images of gastric cancer are classified for cancer detection using a pipeline based on convolutional networks.
Tumour regions are fist classified via patch classification, and a random forest classifier is trained to predict a label at image level, based on hand-crafted features extracted fro... | 3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Whole Slide Image Classification of Gastric Cancer using Convolutional Neural Networks
### Paper Abstract
Gastric cancer is one of the main causes of cancer and cancer-r... |
SJgBra4YDS | ICLR.cc/2020/Conference | 2020 | Manifold Modeling in Embedded Space: A Perspective for Interpreting "Deep Image Prior" | ["Tatsuya Yokota", "Hidekata Hontani", "Qibin Zhao", "Andrzej Cichocki"] | Deep image prior (DIP), which utilizes a deep convolutional network (ConvNet) structure itself as an image prior, has attracted huge attentions in computer vision community. It empirically shows the effectiveness of ConvNet structure for various image restoration applications. However, why the DIP works so well is st... | ["Deep image prior", "Manifold model", "Auto-encoder", "Convolutional neural network", "Delay-embedding", "Hankelization", "Tensor completion", "Image inpainting", "Supperresolution"] | ABSTRACTDeep image prior (DIP) (Ulyanov et al., 2018), which utilizes a deep convolutionalnetwork (ConvNet) structure itself as an image prior, has attracted huge attentionsin computer vision community. It empirically shows the effectiveness of ConvNetstructure for various image restoration applications. However, why t... | S1g1383viS | Official Blind Review #4 | 6: Weak Accept | This paper introduces a transformation from the deep image prior (DIP) to an embedding with an autoencoder (MMES). The authors aim to use this transformation to explain ("in words") why the DIP works so well and explain why convolutions are needed in the DIP. The contributions are summarised as a) providing an interpre... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Manifold Modeling in Embedded Space: A Perspective for Interpreting "Deep Image Prior"
### Paper Abstract
Deep image prior (DIP), which utilizes a deep convolutional net... | |
lFSZySpXXX8 | ICLR.cc/2021/Conference | 2021 | Toward Understanding Supervised Representation Learning with RKHS and GAN | ["Xu Liao", "Jin Liu", "Tianwen Wen", "Yuling Jiao", "Jian Huang"] | The success of deep supervised learning depends on its automatic data representation abilities. A good representation of high-dimensional complex data should enjoy low-dimensionally and disentanglement while losing as little information as possible.
In this work, we give a statistical understanding of how deep represe... | ["Reproducing kernel Hilbert space", "GAN", "neural network", "statistical guarantee"] | ABSTRACTThe success of deep supervised learning depends on its automatic data represen-tation abilities. A good representation of high-dimensional complex data shouldenjoy low-dimensionally and disentanglement while losing as little information aspossible. This work gives a statistical understanding of how a deep repre... | UlcqCQ3embo | REPRESENTATION LEARNING WITH RKHS AND GAN | 5: Marginally below acceptance threshold |
##########################################################################
Summary:
The paper provides an interesting explanation in the representation learning field. In particular, it formulates representation learning as a nonlinear map which characterizes information preservation and disentanglement. To capture ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Toward Understanding Supervised Representation Learning with RKHS and GAN
### Paper Abstract
The success of deep supervised learning depends on its automatic data repres... |
SJlpy64tvB | ICLR.cc/2020/Conference | 2020 | Attacking Lifelong Learning Models with Gradient Reversion | ["Yunhui Guo", "Mingrui Liu", "Yandong Li", "Liqiang Wang", "Tianbao Yang", "Tajana Rosing"] | Lifelong learning aims at avoiding the catastrophic forgetting problem of traditional supervised learning models. Episodic memory based lifelong learning methods such as A-GEM (Chaudhry et al., 2018b) are shown to achieve the state-of-the-art results across the benchmarks. In A-GEM, a small episodic memor... | ["lifelong learning", "adversarial learning"] | ABSTRACTLifelong learning aims at avoiding the catastrophic forgetting problem of tra-ditional supervised learning models. Episodic memory based lifelong learningmethods such as A-GEM (Chaudhry et al., 2018b) are shown to achieve the state-of-the-art results across the benchmarks. In A-GEM, a small episodic memory isut... | B1x9LHd6tr | Official Blind Review #2 | 3: Weak Reject | The paper proposed a novel approach for robust continual learning model from the adversarial attack. The authors start from one of the state-of-the-art episodic memory based continual learning method, A-GEM. The proposed method, Gradient Reversion (GREV), is specialized on A-GEM. The method perturbed the episodic memor... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Attacking Lifelong Learning Models with Gradient Reversion
### Paper Abstract
Lifelong learning aims at avoiding the catastrophic forgetting problem of traditi... | |
SkF2D7g0b | ICLR.cc/2018/Conference | 2018 | Exploring the Space of Black-box Attacks on Deep Neural Networks | ["Arjun Nitin Bhagoji", "Warren He", "Bo Li", "Dawn Song"] | Existing black-box attacks on deep neural networks (DNNs) so far have largely focused on transferability, where an adversarial instance generated for a locally trained model can “transfer” to attack other learning models. In this paper, we propose novel Gradient Estimation black-box attacks for adversaries with query a... | ["adversarial machine learning", "black-box attacks"] | ABSTRACTExisting black-box attacks on deep neural networks (DNNs) so far have largelyfocused on transferability, where an adversarial instance generated for a locallytrained model can “transfer” to attack other learning models. In this paper, wepropose novel Gradient Estimation black-box attacks for adversaries with qu... | B10Nn-jlf | A very thorough empirical examination of practical black-box attacks on NN classifiers. | 7: Good paper, accept | The authors consider new attacks for generating adversarial samples against neural networks. In particular, they are interested in approximating gradient-based white-box attacks such as FGSM in a black-box setting by estimating gradients from queries to the classifier. They assume that the attacker is able to query, fo... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Exploring the Space of Black-box Attacks on Deep Neural Networks
### Paper Abstract
Existing black-box attacks on deep neural networks (DNNs) so far have largely focused... |
B1eXvyHKwS | ICLR.cc/2020/Conference | 2020 | THE EFFECT OF ADVERSARIAL TRAINING: A THEORETICAL CHARACTERIZATION | ["Mingyang Yi", "Huishuai Zhang", "Wei Chen", "Zhi-Ming Ma", "Tie-Yan Liu"] | It has widely shown that adversarial training (Madry et al., 2018) is effective in defending adversarial attack empirically. However, the theoretical understanding of the difference between the solution of adversarial training and that of standard training is limited. In this paper, we characterize the solution of adve... | ["adversarial training", "robustness", "separable data"] | ABSTRACTIt has widely shown that adversarial training (Madry et al., 2018) is effective indefending adversarial attack empirically. However, the theoretical understandingof the difference between the solution of adversarial training and that of stan-dard training is limited. In this paper, we characterize the solution ... | BygvfpMCFS | Official Blind Review #2 | 1: Reject | This paper provides some analyses of the difference between adversarial training and standard training for linear classification problem. In particular, it proves that when the data is \eps linearly separable, adversarial training converges faster than standard trading. It also argues that when the data is not \eps lin... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
THE EFFECT OF ADVERSARIAL TRAINING: A THEORETICAL CHARACTERIZATION
### Paper Abstract
It has widely shown that adversarial training (Madry et al., 2018) is effective in ... | |
LtosxyjROef | CUHK.edu.hk/2021/Course/IERG5350 | 2021 | Reinforcement Learning Based Adaptive WalkingAssistance Control of a Lower Limb Exoskeleton | ["Yingying WANG", "Yutian SHEN"] | Lower limb exoskeleton for people suffering from unilateral weakness or even hemiparesis has attracted considerable interest in recent years. For walking assistance in such scenarios, the exoskeleton is expected to help generate symmetrically coordinated motion with the unaffected side. However, control strategies rema... | ["reinforcement", "adaptive walkingassistance control", "lower limb exoskeleton", "assistance", "system", "limb exoskeleton", "people", "unilateral weakness", "hemiparesis"] | Reinforcement Learning Based Adaptive WalkingAssistance Control of a Lower Limb ExoskeletonYingying Wang, Yutian ShenDepartment of Electronic EngineeringThe Chinese University of Hong KongShatin, Hong Kong{1155135738, 1155113996}@link.cuhk.edu.hkAbstractLower limb exoskeleton for people suffering from unilateral weakne... | amSLsoOx8xX | Creative work, but there are some structure problems of the paper. | 7: Good paper, accept | Generally, this work is hard and creative, it uses different RL methods to control walking assistance in a single-side lower-limb
exoskeleton system. And the paper tests some RL algorithms' performance and adapts some tricks to the optimization of the problem. The simulation performs pretty good. However, there are sti... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Reinforcement Learning Based Adaptive WalkingAssistance Control of a Lower Limb Exoskeleton
### Paper Abstract
Lower limb exoskeleton for people suffering from unilatera... |
tkra4vFiFq | ICLR.cc/2021/Conference | 2021 | GINN: Fast GPU-TEE Based Integrity for Neural Network Training | ["Aref Asvadishirehjini", "Murat Kantarcioglu", "Bradley A. Malin"] | Machine learning models based on Deep Neural Networks (DNNs) are increasingly being deployed in a wide range of applications ranging from self-driving cars to Covid-19 diagnostics. The computational power necessary to learn a DNN is non-trivial. So, as a result, cloud environments with dedicated hardware support emerge... | ["Deep Learning", "Trusted Execution Environments", "Integrity-Preserving Computation", "Intel SGX"] | ABSTRACTMachine learning models based on Deep Neural Networks (DNNs) are increas-ingly deployed in a wide range of applications, ranging from self-driving cars toCOVID-19 treatment discovery. To support the computational power necessary tolearn a DNN, cloud environments with dedicated hardware support have emergedas cr... | BmaRCq7TxrS | The paper targets security challenges of deep neural networks. The authors provide evaluation results to demonstrate their work but more details are needed in some discussions. | 6: Marginally above acceptance threshold | The paper targets security challenges of deep neural networks. While solutions can hardly scale up to support realistic DNN model training workloads, the authors propose GINN to support integrity-preserving DL training by random verification of stochastic gradient steps inside trusted execution environments (TEE). GINN... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
GINN: Fast GPU-TEE Based Integrity for Neural Network Training
### Paper Abstract
Machine learning models based on Deep Neural Networks (DNNs) are increasingly being dep... |
S1jqMb2oM | MIDL.amsterdam/2018/Conference | 2018 | Robust training of recurrent neural networks to handle missing data for disease progression modeling | ["Mostafa Mehdipour Ghazi", "Mads Nielsen", "Akshay Pai", "M. Jorge Cardoso", "Marc Modat", "Sebastien Ourselin", "Lauge S\u00f8rensen"] | Disease progression modeling (DPM) using longitudinal data is a challenging task in machine learning for healthcare that can provide clinicians with better tools for diagnosis and monitoring of disease. Existing DPM algorithms neglect temporal dependencies among measurements and make parametric assumptions about biomar... | ["recurrent neural networks", "long short-term memory", "disease progression modeling", "magnetic resonance imaging", "linear discriminant analysis"] | Robust training of recurrent neural networks tohandle missing data for disease progression modelingMostafa Mehdipour Ghazi1,2,3, Mads Nielsen1,2, Akshay Pai1,2, M. Jorge Cardoso3,Marc Modat3,Sebastien Ourselin3,Lauge Sørensen1,21Biomediq A/S, Copenhagen, DK2Department of Computer Science, Univeristy of Copenhagen, DK3C... | ryvh4M10z | A nice manuscript needing further refinement | 2: Marginally below acceptance threshold | This paper presents an RNN which appears more agnostic to missing data than standard practice.
While the paper is easy to follow, there are a number of points which weren’t addressed reducing the potential impact of the manuscript:
“Finally, subjects with less than three distinct visits per biomarker are removed to be... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust training of recurrent neural networks to handle missing data for disease progression modeling
### Paper Abstract
Disease progression modeling (DPM) using longitud... |
rkx6MJSFPH | ICLR.cc/2020/Conference | 2020 | Unrestricted Adversarial Attacks For Semantic Segmentation | ["Guangyu Shen", "Chengzhi Mao", "Junfeng Yang", "Baishakhi Ray"] | Despite the rapid development of adversarial attacks for machine learning models, many types of new adversarial examples still remain unknown. Uncovered types of adversarial attacks pose serious concern for the safety of the models, which raise the question about the effectiveness of current adversarial robustness ev... | ["Adversarial Attacks", "Semantic Segmentation", "Computer Vision"] | ABSTRACTDespite the rapid development of adversarial attacks for machine learning models,many types of new adversarial examples still remain unknown. Uncovered typesof adversarial attacks pose serious concern for the safety of the models, whichraise the question about the effectiveness of current adversarial robustness... | rkliW9CjtB | Official Blind Review #2 | 6: Weak Accept | This paper uses GANs to generate unrestricted adversarial examples for the task of semantic segmentation, which is traditionally harder to attack via traditional methods (like norm-bounded perturbations). They add an extra term to the GAN loss function for the SPADE generative model. The authors show through experiment... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Unrestricted Adversarial Attacks For Semantic Segmentation
### Paper Abstract
Despite the rapid development of adversarial attacks for machine learning models, many type... | |
q-qxdClTs0d | ICLR.cc/2021/Conference | 2021 | Out-of-distribution Prediction with Invariant Risk Minimization: The Limitation and An Effective Fix | ["Ruocheng Guo", "Pengchuan Zhang", "Hao Liu", "Emre Kiciman"] | This work considers the out-of-distribution (OOD) prediction problem where (1)~the training data are from multiple domains and (2)~the test domain is unseen in the training. DNNs fail in OOD prediction because they are prone to pick up spurious correlations. Recently, Invariant Risk Minimization (IRM) is proposed to ad... | ["Invariant Risk Minimization", "Causal Machine Learning", "Out-of-distribution Prediction"] | ABSTRACTThis work considers the out-of-distribution (OOD) prediction problem where(1) the training data are from multiple domains and (2) the test domain is un-seen in the training. DNNs fail in OOD prediction because they are prone to pickup spurious correlations. Recently, Invariant Risk Minimization (IRM) is pro-pos... | -CseCzHzPB | Paper proposes a new dataset extending CMNIST to highlight flaws of IRM and proposes a fix. Existing works that extend IRM already suffice to fix this problem. Several theoretical issues with fix proposed by the authors. | 4: Ok but not good enough - rejection | Summary:
In this work, the authors focus on the out-of-distribution generalization problem. The input is dataset from multiple environments and the goal is to learn a model that generalizes well to an unseen test environment. The work is based on recent line of works on invariant risk minimization (IRM) (Arjovsky et al... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Out-of-distribution Prediction with Invariant Risk Minimization: The Limitation and An Effective Fix
### Paper Abstract
This work considers the out-of-distribution (OOD)... |
HkgmzhC5F7 | ICLR.cc/2019/Conference | 2019 | A Modern Take on the Bias-Variance Tradeoff in Neural Networks | ["Brady Neal", "Sarthak Mittal", "Aristide Baratin", "Vinayak Tantia", "Matthew Scicluna", "Simon Lacoste-Julien", "Ioannis Mitliagkas"] | We revisit the bias-variance tradeoff for neural networks in light of modern empirical findings. The traditional bias-variance tradeoff in machine learning suggests that as model complexity grows, variance increases. Classical bounds in statistical learning theory point to the number of parameters in a model as a measu... | ["bias-variance tradeoff", "deep learning theory", "generalization", "concentration"] | ABSTRACTWe revisit the bias-variance tradeoff for neural networks in light of modern empir-ical findings. The traditional bias-variance tradeoff in machine learning suggeststhat as model complexity grows, variance increases. Classical bounds in statis-tical learning theory point to the number of parameters in a model a... | SJlSHgE03Q | Interesting paper with some experiments and preliminary results but requires more work | 5: Marginally below acceptance threshold | This paper studies variance-bias tradeoff as a function of depth and width of a neural network. Experiments suggest that variance may decrease as a function width and increase as a function of depth. Some analytical results are presented why this may the case for width and why the necessary assumptions for the depth ar... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Modern Take on the Bias-Variance Tradeoff in Neural Networks
### Paper Abstract
We revisit the bias-variance tradeoff for neural networks in light of modern empirical ... |
H1gKYo09tX | ICLR.cc/2019/Conference | 2019 | code2seq: Generating Sequences from Structured Representations of Code | ["Uri Alon", "Shaked Brody", "Omer Levy", "Eran Yahav"] | The ability to generate natural language sequences from source code snippets has a variety of applications such as code summarization, documentation, and retrieval. Sequence-to-sequence (seq2seq) models, adopted from neural machine translation (NMT), have achieved state-of-the-art performance on these tasks by treating... | ["source code", "programs", "code2seq"] | ABSTRACTThe ability to generate natural language sequences from source code snippets hasa variety of applications such as code summarization, documentation, and re-trieval. Sequence-to-sequence (seq2seq) models, adopted from neural machinetranslation (NMT), have achieved state-of-the-art performance on these tasks bytr... | BJgTXkO53Q | Nice model using path summaries, but some baseline comparisons are missing | 5: Marginally below acceptance threshold | This paper presents a new code-to-sequence model called code2seq that leverages the syntactic structure of programming languages to encode source code snippets, which is then decoded to natural language using a sequence decoder. The key idea of the approach is to represent a program using a set of randomly sample k pat... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
code2seq: Generating Sequences from Structured Representations of Code
### Paper Abstract
The ability to generate natural language sequences from source code snippets ha... |
6SXNhWc5HFe | ICLR.cc/2021/Conference | 2021 | Provable Fictitious Play for General Mean-Field Games | ["Qiaomin Xie", "Zhuoran Yang", "Zhaoran Wang", "Andreea Minca"] | We propose a reinforcement learning algorithm for stationary mean-field games, where the goal is to learn a pair of mean-field state and stationary policy that constitutes the Nash equilibrium. When viewing the mean-field state and the policy as two players, we propose a fictitious play algorithm which alternatively up... | ["Mean-field games", "Fictitious play", "Entropy regularization", "Nash equilibrium"] | ABSTRACTWe propose a reinforcement learning algorithm for stationary mean-field games,where the goal is to learn a pair of mean-field state and stationary policy that con-stitutes the Nash equilibrium. When viewing the mean-field state and the policyas two players, we propose a fictitious play algorithm which alternati... | VlgAcA-IDtI | Overall solid work but with unclear novelty, that needs to be better motivated | 5: Marginally below acceptance threshold | The paper considers a Markov mean-field game where the continuum of players is summarized by their distribution over the state space. The policy of each player is a function of the state alone (and not a function of mean-field distribution). It is assumed that the game is "well-behaved" so it can be embedded in a regul... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Provable Fictitious Play for General Mean-Field Games
### Paper Abstract
We propose a reinforcement learning algorithm for stationary mean-field games, where the goal is... |
SkJdByaUG | ICLR.cc/2018/Workshop | 2018 | Forecasting probability distribution of nonlinear time series | ["Kyongmin Yeo", "Igor Melnyk", "Nam Nguyen", "Eun Kyung Lee"] | We propose DE-RNN to learn the probability density function (PDF) of a nonlinear time series, and compute the temporal evolution of the PDF for a probabilistic forecast. A Recurrent Neural Network (RNN) based model is employed to learn a nonlinear operator for temporal evolution of the stochastic process. We use a soft... | ["Recurrent Neural Network", "forecast", "density estimation", "nonlinear time series", "dynamical system", "uncertainty quantification"] | ABSTRACTWe propose DE-RNN to learn the probability density function (PDF) of a nonlin-ear time series, and compute the temporal evolution of the PDF for a probabilisticforecast. A Recurrent Neural Network (RNN) based model is employed to learn anonlinear operator for temporal evolution of the stochastic process. We use... | HJB7FSMtM | Interesting work which could be potentially improved | 5: Marginally below acceptance threshold | This paper seeks to solve time-series PDF estimation problem. It applies a RNN to capture the temporal dependency over the past observations. It also uses a grid discretization to transform a PDF estimation formulation into a classification problem. Experiments show that the proposed approach perform favorably compared... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Forecasting probability distribution of nonlinear time series
### Paper Abstract
We propose DE-RNN to learn the probability density function (PDF) of a nonlinear time se... |
q1o2mWaOssG | NeurIPS.cc/2020/Workshop/SVRHM | 2020 | Brain-inspired predictive coding dynamics improve the robustness of deep neural networks | ["Bhavin Choksi", "Milad Mozafari", "Callum Biggs O'May", "B. ADOR", "Andrea Alamia", "Rufin VanRullen"] | Deep neural networks excel at image classification, but their performance is far less robust to input perturbations than human perception. In this work we address this shortcoming by incorporating brain-inspired recurrent dynamics in deep convolutional networks. We augment a pretrained feedforward classification model ... | ["predictive coding", "neuroscience", "robustness", "machine learning", "deep learning"] | Brain-inspired predictive coding dynamics improvethe robustness of deep neural networksBhavin ChoksiCerCo CNRS, UMR 5549 &Université de Toulousebhavin.choksi@cnrs.frMilad Mozafari*CerCo CNRS, UMR 5549 &IRIT CNRS, UMR 5505milad.mozafari@cnrs.frCallum Biggs O’MayCerCo CNRSUMR 5549Benjamin AdorCerCo CNRSUMR 5549Andrea Al... | HLDVCpbhmdv | Important problem and interesting results, however unclear impact on traditional classification performance | 7: Good paper, accept | The problem that the paper addresses is an important and an interesting one. The role of feedback in networks is still disputed and this paper provides some evidence towards a solution. The approach is fairly straightforward yet provides some interesting results. Though there are still many open questions remaining, su... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Brain-inspired predictive coding dynamics improve the robustness of deep neural networks
### Paper Abstract
Deep neural networks excel at image classification, but their... |
ryeUg0VFwr | ICLR.cc/2020/Conference | 2020 | Striving for Simplicity in Off-Policy Deep Reinforcement Learning | ["Rishabh Agarwal", "Dale Schuurmans", "Mohammad Norouzi"] | This paper advocates the use of offline (batch) reinforcement learning (RL) to help (1) isolate the contributions of exploitation vs. exploration in off-policy deep RL, (2) improve reproducibility of deep RL research, and (3) facilitate the design of simpler deep RL algorithms. We propose an offline RL benchmark on Ata... | ["reinforcement learning", "off-policy", "batch RL", "offline RL", "benchmark"] | ABSTRACTThis paper advocates the use of offline (batch) reinforcement learning (RL) tohelp (1) isolate the contributions of exploitation vs. exploration in off-policydeep RL, (2) improve reproducibility of deep RL research, and (3) facilitate thedesign of simpler deep RL algorithms. We demonstrate that recent off-polic... | BylZbsQTYB | Official Blind Review #1 | 3: Weak Reject | This paper investigated the offline settings of deep Q-networks, and provided relevant comparison among different agents. The authors also proposed two DQN variants by using ensembles, and showed the improvement (mainly from the random ensemble mixture) on Atari games, when compared with baselines.
In general, this p... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Striving for Simplicity in Off-Policy Deep Reinforcement Learning
### Paper Abstract
This paper advocates the use of offline (batch) reinforcement learning (RL) to help ... | |
rklv-a4tDB | ICLR.cc/2020/Conference | 2020 | Mesh-Free Unsupervised Learning-Based PDE Solver of Forward and Inverse problems | ["Leah Bar", "Nir Sochen"] | We introduce a novel neural network-based partial differential equations solver for forward and inverse problems. The solver is grid free, mesh free and shape free, and the solution is approximated by a neural network.
We employ an unsupervised approach such that the input to the network is a points set in an arbitrar... | ["PDEs", "forward problems", "inverse problems", "unsupervised learning", "deep networks", "EIT"] | ABSTRACTWe introduce a novel neural network-based partial differential equations solverfor forward and inverse problems. The solver is grid free, mesh free and shapefree, and the solution is approximated by a neural network. We employ an unsu-pervised approach such that the input to the network is a points set in an ar... | Hyg9zSZCYr | Official Blind Review #3 | 3: Weak Reject | This paper presents an unsupervised learning approach to solve forward and inverse problems represented by partial differential equations. A framework is proposed based on the minimisation of loss functions that enforce the boundary conditions of the PDE solution and promote its smoothness. The method is then applied t... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Mesh-Free Unsupervised Learning-Based PDE Solver of Forward and Inverse problems
### Paper Abstract
We introduce a novel neural network-based partial differential equati... | |
HyzsgBEtg | ICLR.cc/2017/workshop | 2017 | Dance Dance Convolution | ["Chris Donahue", "Zachary C. Lipton", "Julian McAuley"] | Dance Dance Revolution (DDR) is a popular rhythm-based video game. Players perform steps on a dance platform in synchronization with music as directed by on-screen step charts. While many step charts are available in standardized packs, users may grow tired of existing charts, or wish to dance to a song for which no ch... | ["Deep learning", "Supervised Learning", "Applications", "Games"] | ABSTRACTDance Dance Revolution (DDR) is a popular rhythm-based video game. Playersperform steps on a dance platform in synchronization with music as directed byon-screen step charts . While many step charts are available in standardized packs,users may grow tired of existing charts, or wish to dance to a song for which... | r1yGGiHje | an interesting application and a work in progress | 6: Marginally above acceptance threshold | This is an interesting application of DL in rhythm-based video games that learns two sub-tasks: step placement from raw audio, and step selection from ground truth placement. This seems to be a work in progress, as it doesn't address the complete end-to-end problem yet.
Some key information is missing in the paper. F... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Dance Dance Convolution
### Paper Abstract
Dance Dance Revolution (DDR) is a popular rhythm-based video game. Players perform steps on a dance platform in synchronizatio... |
bSpvqVACDmQ | MLSys/2023/Workshop/RCLWN | 2023 | DHA-FL: Enabling Efficient and Effective AIoT via Decentralized Hierarchical Asynchronous Federated Learning | ["Wesley Houston Huff", "pinyarash pinyoanuntapong", "Ravikumar Balakrishnan", "Hao Feng", "Minwoo Lee", "Pu Wang", "Chen Chen"] | The challenges of scalability, robustness, and resilience to slow devices have posed significant obstacles to the effective and efficient implementation of Federated Learning (FL), a crucial technology for the emerging Artificial Intelligence of Things (AIoT). This paper proposes a solution to these challenges with the... | ["Machine Learning", "MLSys"] | ABSTRACTThe challenges of scalability, robustness, and resilience to slow devices have posed significant obstacles to theeffective and efficient implementation of Federated Learning (FL), a crucial technology for the emerging Ar-tificial Intelligence of Things (AIoT). This paper proposes a solution to these challenges ... | OklPJE7facB | Official Review of Paper3 by Reviewer 7xTR | 4: Ok but not good enough - rejection | The paper presents a decentralized hierarchical federated learning scheme aimed at decreasing the convergence time of federated learning when devices are heterogeneous. The proposed DHA-FL scheme consists of two layers: edge servers and workers. In each training cycle, workers update their model and share it with thei... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
DHA-FL: Enabling Efficient and Effective AIoT via Decentralized Hierarchical Asynchronous Federated Learning
### Paper Abstract
The challenges of scalability, robustness... |
4x2RUQ99sGz | robot-learning.org/CoRL/2023/Conference | 2023 | Online Model Adaptation with Feedforward Compensation | ["ABULIKEMU ABUDUWEILI", "Changliu Liu"] | To cope with distribution shifts or non-stationarity in system dynamics, online adaptation algorithms have been introduced to update offline-learned prediction models in real-time. Existing online adaptation methods focus on optimizing the prediction model by utilizing feedback from the latest prediction error. Unfort... | ["Online Adaptation", "Optimization", "Behavior prediction"] | Online Model Adaptationwith Feedforward CompensationAbulikemu Abuduweili, Changliu LiuRobotics Institute, Carnegie Mellon University{abulikea,cliu6 }@andrew.cmu.eduAbstract: To cope with distribution shifts or non-stationarity in system dynam-ics, online adaptation algorithms have been introduced to update offline-lear... | isDSHKGvd2 | quality: It is a high quality paper and I enjoyed reading it
clarity: The paper is well written and the presented method is clear
originality: The idea seems to be original and, as far as I know, a new and interesting way to see the problem of online model adaptation
significance: As the general pipeline is not new ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Online Model Adaptation with Feedforward Compensation
### Paper Abstract
To cope with distribution shifts or non-stationarity in system dynamics, online adaptation algor... | ||
S1fNJhRqFX | ICLR.cc/2019/Conference | 2019 | Exploration using Distributional RL and UCB | ["Borislav Mavrin", "Hengshuai Yao", "Linglong Kong", "ShangtongZhang"] | We establish the relation between Distributional RL and the Upper Confidence Bound (UCB) approach to exploration.
In this paper we show that the density of the Q function estimated by Distributional RL can be successfully used for the estimation of UCB. This approach does not require counting and, therefore, ge... | ["Distributional RL", "UCB", "exploration", "Atari 2600", "multi-armed bandits"] | ABSTRACTWe establish the relation between Distributional RL and the Upper ConfidenceBound (UCB) approach to exploration. In this paper we show that the density ofthe Q function estimated by Distributional RL can be successfully used for theestimation of UCB. This approach does not require counting and, therefore, gen-e... | BkxGNwYV3X | The reviewer does not understand why equation 2 solves the exploration-exploitation tradeoff. | 4: Ok but not good enough - rejection | This paper proposes new algorithms (QUCB and QUCB+) to handle the exploration exploitation tradeoff in Multi-Armed Bandits and more generally in Reinforcement Learning. The first idea of the paper is to use an Upper Confidence Bound based on the variance estimation rather than an exploration factor based on a concentra... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Exploration using Distributional RL and UCB
### Paper Abstract
We establish the relation between Distributional RL and the Upper Confidence Bound (UCB) approach to e... |
pWipslK5xVf | ICLR.cc/2021/Conference | 2021 | Wasserstein Distributional Normalization : Nonparametric Stochastic Modeling for Handling Noisy Labels | ["Sung Woo Park", "Junseok Kwon"] | We propose a novel Wasserstein distributional normalization (WDN) algorithm to handle noisy labels for accurate classification. In this paper, we split our data into uncertain and certain samples based on small loss criteria. We investigate the geometric relationship between these two different types of samples and en... | ["Wasserstein distributional normalization", "Noisy labels", "Classification"] | ABSTRACTWe propose a novel Wasserstein distributional normalization (WDN) algorithmto handle noisy labels for accurate classification. In this paper, we split our datainto uncertain and certain samples based on small loss criteria. We investigate thegeometric relationship between these two different types of samples an... | ew0pxf0UC8X | Recommendation to Accept | 6: Marginally above acceptance threshold | This paper propose a computationally efficient Wasserstein distributional normalization algorithm for accurate classification of noisy labels. An explicit upper bound for the Wasserstein-2 distance is derived and such a bound can be used as an estimator to determine if a network is over-parameterized. Empirical results... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Wasserstein Distributional Normalization : Nonparametric Stochastic Modeling for Handling Noisy Labels
### Paper Abstract
We propose a novel Wasserstein distributional n... |
SkxwY3NtPr | ICLR.cc/2020/Conference | 2020 | Towards a Unified Evaluation of Explanation Methods without Ground Truth | ["Hao Zhang", "Jiayi Chen", "Haotian Xue", "Quanshi Zhang"] | This paper proposes a set of criteria to evaluate the objectiveness of explanation
methods of neural networks, which is crucial for the development of explainable
AI, but it also presents significant challenges. The core challenge is that people
usually cannot obtain ground-truth explanations of the neural network. To ... | ["Deep Learning", "Interpretability", "Evaluation", "Convolutional Neural Networks"] | ABSTRACTThis paper proposes a set of criteria to evaluate the objectiveness of explanationmethods of neural networks, which is crucial for the development of explainableAI, but it also presents significant challenges. The core challenge is that peopleusually cannot obtain ground-truth explanations of the neural network... | rkg04TanFH | Official Blind Review #1 | 6: Weak Accept |
The paper proposes four metrics to evaluate explanation methods, namely 1) bias of the attribution map based on Shapley value, 2) unexplainable feature components, 3) robustness of explanations, and 4) mutual verification (whether two explanation methods can verify each other). These metrics focus on explanation metho... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards a Unified Evaluation of Explanation Methods without Ground Truth
### Paper Abstract
This paper proposes a set of criteria to evaluate the objectiveness of explan... | |
mCLVeEpplNE | ICLR.cc/2021/Conference | 2021 | NBDT: Neural-Backed Decision Tree | ["Alvin Wan", "Lisa Dunlap", "Daniel Ho", "Jihan Yin", "Scott Lee", "Suzanne Petryk", "Sarah Adel Bargal", "Joseph E. Gonzalez"] | Machine learning applications such as finance and medicine demand accurate and justifiable predictions, barring most deep learning methods from use. In response, previous work combines decision trees with deep learning, yielding models that (1) sacrifice interpretability for accuracy or (2) sacrifice accuracy for inter... | ["explainability", "computer vision", "interpretability"] | ABSTRACTMachine learning applications such as finance and medicine demand accurate andjustifiable predictions, barring most deep learning methods from use. In response,previous work combines decision trees with deep learning, yielding models that(1) sacrifice interpretability for accuracy or (2) sacrifice accuracy for ... | ITf63wjbo4F | The paper presents an algorithm (NBDT) for interpretable image classification, demonstrating a higher preference rate (65.9%) when compared to a saliency map for the model explanation among 374 participants. | 6: Marginally above acceptance threshold | Significance:
This article seems to be a useful contribution to the literature on interpretable deep networks. However, the paper could be strengthened by demonstrating and analyzing the interpretability of approaches to other types of data, such as sequential data.
Novelty:
This paper's main contribution is to offer ... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NBDT: Neural-Backed Decision Tree
### Paper Abstract
Machine learning applications such as finance and medicine demand accurate and justifiable predictions, barring most... |
HkGGfhC5Y7 | ICLR.cc/2019/Conference | 2019 | Towards a better understanding of Vector Quantized Autoencoders | ["Aurko Roy", "Ashish Vaswani", "Niki Parmar", "Arvind Neelakantan"] | Deep neural networks with discrete latent variables offer the promise of better symbolic reasoning, and learning abstractions that are more useful to new tasks. There has been a surge in interest in discrete latent variable models, however, despite several recent improvements, the training of discrete latent variable... | ["machine translation", "vector quantized autoencoders", "non-autoregressive", "NMT"] | Under review as a conference paper at ICLR 2019Towards a better understanding of VectorQuantized AutoencodersAnonymous authorsPaper under double-blind reviewAbstractDeep neural networks with discrete latent variables offer the promise ofbetter symbolic reasoning, and learning abstractions that are more usefulto new tas... | r1lCtgTx27 | Experimental section | 6: Marginally above acceptance threshold | This paper introduces a new way of interpreting the VQ-VAE,
and proposes a new training algorithm based on the soft EM clustering.
I think the technical aspect of this paper is written concisely.
Introducing the interpretation as hard EM seems natural for me, and the extension
to the soft EM training is sound reaso... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards a better understanding of Vector Quantized Autoencoders
### Paper Abstract
Deep neural networks with discrete latent variables offer the promise of better symbol... |
ByxGkySKwH | ICLR.cc/2020/Conference | 2020 | Towards neural networks that provably know when they don't know | ["Alexander Meinke", "Matthias Hein"] | It has recently been shown that ReLU networks produce arbitrarily over-confident predictions far away from the
training data. Thus, ReLU networks do not know when they don't know. However, this is a highly important property in safety
critical applications. In the context of out-of-distribution detection (OOD) there h... | ["relu networks", "towards neural networks", "training data", "low confidence predictions", "predictions", "important property", "safety critical applications", "context", "detection", "ood"] | ABSTRACTIt has recently been shown that ReLU networks produce arbitrarily over-confidentpredictions far away from the training data. Thus, ReLU networks do not knowwhen they don’t know. However, this is a highly important property in safetycritical applications. In the context of out-of-distribution detection (OOD) the... | BkxIutEqtS | Official Blind Review #1 | 6: Weak Accept | The authors present a novel approach for OOD detection; in particular their approach comes with worst-case guarantees without compromising on performance.
The manuscript is clearly written and I only have some concerns regarding the evaluation.
First, while the authors include with MCD an uncertainty-aware training ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards neural networks that provably know when they don't know
### Paper Abstract
It has recently been shown that ReLU networks produce arbitrarily over-confident predi... | |
bgQek2O63w | ICLR.cc/2021/Conference | 2021 | Self-supervised Adversarial Robustness for the Low-label, High-data Regime | ["Sven Gowal", "Po-Sen Huang", "Aaron van den Oord", "Timothy Mann", "Pushmeet Kohli"] | Recent work discovered that training models to be invariant to adversarial perturbations requires substantially larger datasets than those required for standard classification. Perhaps more surprisingly, these larger datasets can be "mostly" unlabeled. Pseudo-labeling, a technique simultaneously pioneered by four separ... | ["self-supervised", "adversarial training", "robustness"] | ABSTRACTRecent work discovered that training models to be invariant to adversarial per-turbations requires substantially larger datasets than those required for standardclassification. Perhaps more surprisingly, these larger datasets can be “mostly”unlabeled. Pseudo-labeling , a technique simultaneously pioneered by fo... | RCQOECCAsis | Application based approach, claims not very well supported. | 4: Ok but not good enough - rejection |
This paper proposes some modifications to BYOL (Bootstrap Your Own Latents) in an attempt to address adversarial robustness in low-label high data regimes. The paper is well written and very easy to follow. Overall, the idea is clear and well presented.
Although the author's claim their approach to be novel, as is, ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Self-supervised Adversarial Robustness for the Low-label, High-data Regime
### Paper Abstract
Recent work discovered that training models to be invariant to adversarial ... |
Bkv9FyHYx | ICLR.cc/2017/workshop | 2017 | Joint Training of Ratings and Reviews with Recurrent Recommender Networks | ["Chao-Yuan Wu", "Amr Ahmed", "Alex Beutel", "Alexander J. Smola"] | Accurate modeling of ratings and text reviews is at the core of successful recommender systems. While neural networks have been remarkably successful in modeling images and natural language, they have been largely unexplored in recommender system research. In this paper, we provide a neural network model that combines... | ["ratings", "reviews", "text reviews", "joint training", "recurrent recommender networks", "modeling", "core", "successful recommender systems", "neural networks"] | ABSTRACTAccurate modeling of ratings and text reviews is at the core of successful recom-mender systems. In this paper, we provide a neural network model that combinesratings, reviews, and temporal patterns to learn highly accurate recommendations.We co-train for prediction on both numerical ratings and natural languag... | SyII9Uyog | 10: Top 5% of accepted papers, seminal paper | This paper proposed a joint model for rate prediction and text generation. The author compared the methods on a more realistic time based split setting, which requires “predict into the future.” The model is interesting and sound, and I think it should be accepted to ICLR workshop | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Joint Training of Ratings and Reviews with Recurrent Recommender Networks
### Paper Abstract
Accurate modeling of ratings and text reviews is at the core of successful r... | |
qSeqhriWKsn | ICLR.cc/2021/Conference | 2021 | Adaptive Single-Pass Stochastic Gradient Descent in Input Sparsity Time | ["Sepideh Mahabadi", "David Woodruff", "Samson Zhou"] | We study sampling algorithms for variance reduction methods for stochastic optimization. Although stochastic gradient descent (SGD) is widely used for large scale machine learning, it sometimes experiences slow convergence rates due to the high variance from uniform sampling. In this paper, we introduce an algorithm th... | ["stochastic gradient descent", "streaming algorithm", "stochastic optimization"] | ABSTRACTWe study sampling algorithms for variance reduction methods for stochastic op-timization. Although stochastic gradient descent (SGD) is widely used for largescale machine learning, it sometimes experiences slow convergence rates due tothe high variance from uniform sampling. In this paper, we introduce an algor... | gf0xxD0r-8v | Interesting theoretical application of sketching; clarity issues | 6: Marginally above acceptance threshold | Summary:
This paper develops an efficient streaming algorithm to approximate the optimal importance sampling weights for variance reduction in finite-sum SGD. The optimal weights are proportional to each sample's gradient norm; this work uses AMS-like moment estimation to sketch gradient norms which take the form of a ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adaptive Single-Pass Stochastic Gradient Descent in Input Sparsity Time
### Paper Abstract
We study sampling algorithms for variance reduction methods for stochastic opt... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.